1 //===-- X86InstrSSE.td - SSE Instruction Set ---------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
16 class OpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm> {
17 InstrItinClass rr = arg_rr;
18 InstrItinClass rm = arg_rm;
19 // InstrSchedModel info.
20 X86FoldableSchedWrite Sched = WriteFAdd;
23 class SizeItins<OpndItins arg_s, OpndItins arg_d> {
29 class ShiftOpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm,
30 InstrItinClass arg_ri> {
31 InstrItinClass rr = arg_rr;
32 InstrItinClass rm = arg_rm;
33 InstrItinClass ri = arg_ri;
38 let Sched = WriteFAdd in {
39 def SSE_ALU_F32S : OpndItins<
40 IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM
43 def SSE_ALU_F64S : OpndItins<
44 IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM
48 def SSE_ALU_ITINS_S : SizeItins<
49 SSE_ALU_F32S, SSE_ALU_F64S
52 let Sched = WriteFMul in {
53 def SSE_MUL_F32S : OpndItins<
54 IIC_SSE_MUL_F32S_RR, IIC_SSE_MUL_F64S_RM
57 def SSE_MUL_F64S : OpndItins<
58 IIC_SSE_MUL_F64S_RR, IIC_SSE_MUL_F64S_RM
62 def SSE_MUL_ITINS_S : SizeItins<
63 SSE_MUL_F32S, SSE_MUL_F64S
66 let Sched = WriteFDiv in {
67 def SSE_DIV_F32S : OpndItins<
68 IIC_SSE_DIV_F32S_RR, IIC_SSE_DIV_F64S_RM
71 def SSE_DIV_F64S : OpndItins<
72 IIC_SSE_DIV_F64S_RR, IIC_SSE_DIV_F64S_RM
76 def SSE_DIV_ITINS_S : SizeItins<
77 SSE_DIV_F32S, SSE_DIV_F64S
81 let Sched = WriteFAdd in {
82 def SSE_ALU_F32P : OpndItins<
83 IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM
86 def SSE_ALU_F64P : OpndItins<
87 IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM
91 def SSE_ALU_ITINS_P : SizeItins<
92 SSE_ALU_F32P, SSE_ALU_F64P
95 let Sched = WriteFMul in {
96 def SSE_MUL_F32P : OpndItins<
97 IIC_SSE_MUL_F32P_RR, IIC_SSE_MUL_F64P_RM
100 def SSE_MUL_F64P : OpndItins<
101 IIC_SSE_MUL_F64P_RR, IIC_SSE_MUL_F64P_RM
105 def SSE_MUL_ITINS_P : SizeItins<
106 SSE_MUL_F32P, SSE_MUL_F64P
109 let Sched = WriteFDiv in {
110 def SSE_DIV_F32P : OpndItins<
111 IIC_SSE_DIV_F32P_RR, IIC_SSE_DIV_F64P_RM
114 def SSE_DIV_F64P : OpndItins<
115 IIC_SSE_DIV_F64P_RR, IIC_SSE_DIV_F64P_RM
119 def SSE_DIV_ITINS_P : SizeItins<
120 SSE_DIV_F32P, SSE_DIV_F64P
123 def SSE_BIT_ITINS_P : OpndItins<
124 IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
127 let Sched = WriteVecALU in {
128 def SSE_INTALU_ITINS_P : OpndItins<
129 IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
132 def SSE_INTALUQ_ITINS_P : OpndItins<
133 IIC_SSE_INTALUQ_P_RR, IIC_SSE_INTALUQ_P_RM
137 let Sched = WriteVecIMul in
138 def SSE_INTMUL_ITINS_P : OpndItins<
139 IIC_SSE_INTMUL_P_RR, IIC_SSE_INTMUL_P_RM
142 def SSE_INTSHIFT_ITINS_P : ShiftOpndItins<
143 IIC_SSE_INTSH_P_RR, IIC_SSE_INTSH_P_RM, IIC_SSE_INTSH_P_RI
146 def SSE_MOVA_ITINS : OpndItins<
147 IIC_SSE_MOVA_P_RR, IIC_SSE_MOVA_P_RM
150 def SSE_MOVU_ITINS : OpndItins<
151 IIC_SSE_MOVU_P_RR, IIC_SSE_MOVU_P_RM
154 //===----------------------------------------------------------------------===//
155 // SSE 1 & 2 Instructions Classes
156 //===----------------------------------------------------------------------===//
158 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
159 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
160 RegisterClass RC, X86MemOperand x86memop,
163 let isCommutable = 1 in {
164 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
166 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
167 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
168 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>,
169 Sched<[itins.Sched]>;
171 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
173 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
174 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
175 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>,
176 Sched<[itins.Sched.Folded, ReadAfterLd]>;
179 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
180 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
181 string asm, string SSEVer, string FPSizeStr,
182 Operand memopr, ComplexPattern mem_cpat,
185 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
187 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
188 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
189 [(set RC:$dst, (!cast<Intrinsic>(
190 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
191 RC:$src1, RC:$src2))], itins.rr>,
192 Sched<[itins.Sched]>;
193 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
195 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
196 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
197 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
198 SSEVer, "_", OpcodeStr, FPSizeStr))
199 RC:$src1, mem_cpat:$src2))], itins.rm>,
200 Sched<[itins.Sched.Folded, ReadAfterLd]>;
203 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
204 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
205 RegisterClass RC, ValueType vt,
206 X86MemOperand x86memop, PatFrag mem_frag,
207 Domain d, OpndItins itins, bit Is2Addr = 1> {
208 let isCommutable = 1 in
209 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
211 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
212 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
213 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>,
214 Sched<[itins.Sched]>;
216 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
218 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
219 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
220 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
222 Sched<[itins.Sched.Folded, ReadAfterLd]>;
225 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
226 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
227 string OpcodeStr, X86MemOperand x86memop,
228 list<dag> pat_rr, list<dag> pat_rm,
230 let isCommutable = 1, hasSideEffects = 0 in
231 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
233 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
234 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
235 pat_rr, NoItinerary, d>,
236 Sched<[WriteVecLogic]>;
237 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
239 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
240 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
241 pat_rm, NoItinerary, d>,
242 Sched<[WriteVecLogicLd, ReadAfterLd]>;
245 //===----------------------------------------------------------------------===//
246 // Non-instruction patterns
247 //===----------------------------------------------------------------------===//
249 // A vector extract of the first f32/f64 position is a subregister copy
250 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
251 (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32)>;
252 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
253 (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64)>;
255 // A 128-bit subvector extract from the first 256-bit vector position
256 // is a subregister copy that needs no instruction.
257 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (iPTR 0))),
258 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
259 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (iPTR 0))),
260 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
262 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (iPTR 0))),
263 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
264 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (iPTR 0))),
265 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
267 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (iPTR 0))),
268 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
269 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (iPTR 0))),
270 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
272 // A 128-bit subvector insert to the first 256-bit vector position
273 // is a subregister copy that needs no instruction.
274 let AddedComplexity = 25 in { // to give priority over vinsertf128rm
275 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)),
276 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
277 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)),
278 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
279 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)),
280 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
281 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)),
282 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
283 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (iPTR 0)),
284 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
285 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (iPTR 0)),
286 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
289 // Implicitly promote a 32-bit scalar to a vector.
290 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
291 (COPY_TO_REGCLASS FR32:$src, VR128)>;
292 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
293 (COPY_TO_REGCLASS FR32:$src, VR128)>;
294 // Implicitly promote a 64-bit scalar to a vector.
295 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
296 (COPY_TO_REGCLASS FR64:$src, VR128)>;
297 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
298 (COPY_TO_REGCLASS FR64:$src, VR128)>;
300 // Bitcasts between 128-bit vector types. Return the original type since
301 // no instruction is needed for the conversion
302 let Predicates = [HasSSE2] in {
303 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
304 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
305 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
306 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
307 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
308 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
309 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
310 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
311 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
312 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
313 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
314 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
315 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
316 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
317 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
318 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
319 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
320 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
321 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
322 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
323 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
324 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
325 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
326 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
327 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
328 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
329 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
330 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
331 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
332 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
335 // Bitcasts between 256-bit vector types. Return the original type since
336 // no instruction is needed for the conversion
337 let Predicates = [HasAVX] in {
338 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
339 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
340 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
341 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
342 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
343 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
344 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
345 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
346 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
347 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
348 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
349 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
350 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
351 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
352 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
353 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
354 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
355 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
356 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
357 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
358 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
359 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
360 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
361 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
362 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
363 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
364 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
365 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
366 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
367 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
370 // Alias instructions that map fld0 to xorps for sse or vxorps for avx.
371 // This is expanded by ExpandPostRAPseudos.
372 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
373 isPseudo = 1, SchedRW = [WriteZero] in {
374 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
375 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
376 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
377 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
380 //===----------------------------------------------------------------------===//
381 // AVX & SSE - Zero/One Vectors
382 //===----------------------------------------------------------------------===//
384 // Alias instruction that maps zero vector to pxor / xorp* for sse.
385 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
386 // swizzled by ExecutionDepsFix to pxor.
387 // We set canFoldAsLoad because this can be converted to a constant-pool
388 // load of an all-zeros value if folding it would be beneficial.
389 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
390 isPseudo = 1, SchedRW = [WriteZero] in {
391 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
392 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
395 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
396 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
397 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
398 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
399 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
402 // The same as done above but for AVX. The 256-bit AVX1 ISA doesn't support PI,
403 // and doesn't need it because on sandy bridge the register is set to zero
404 // at the rename stage without using any execution unit, so SET0PSY
405 // and SET0PDY can be used for vector int instructions without penalty
406 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
407 isPseudo = 1, Predicates = [HasAVX], SchedRW = [WriteZero] in {
408 def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
409 [(set VR256:$dst, (v8f32 immAllZerosV))]>;
412 let Predicates = [HasAVX] in
413 def : Pat<(v4f64 immAllZerosV), (AVX_SET0)>;
415 let Predicates = [HasAVX2] in {
416 def : Pat<(v4i64 immAllZerosV), (AVX_SET0)>;
417 def : Pat<(v8i32 immAllZerosV), (AVX_SET0)>;
418 def : Pat<(v16i16 immAllZerosV), (AVX_SET0)>;
419 def : Pat<(v32i8 immAllZerosV), (AVX_SET0)>;
422 // AVX1 has no support for 256-bit integer instructions, but since the 128-bit
423 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
424 let Predicates = [HasAVX1Only] in {
425 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
426 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
427 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
429 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
430 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
431 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
433 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
434 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
435 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
437 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
438 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
439 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
442 // We set canFoldAsLoad because this can be converted to a constant-pool
443 // load of an all-ones value if folding it would be beneficial.
444 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
445 isPseudo = 1, SchedRW = [WriteZero] in {
446 def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
447 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
448 let Predicates = [HasAVX2] in
449 def AVX2_SETALLONES : I<0, Pseudo, (outs VR256:$dst), (ins), "",
450 [(set VR256:$dst, (v8i32 immAllOnesV))]>;
454 //===----------------------------------------------------------------------===//
455 // SSE 1 & 2 - Move FP Scalar Instructions
457 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
458 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
459 // is used instead. Register-to-register movss/movsd is not modeled as an
460 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
461 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
462 //===----------------------------------------------------------------------===//
464 multiclass sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt,
465 X86MemOperand x86memop, string base_opc,
467 def rr : SI<0x10, MRMSrcReg, (outs VR128:$dst),
468 (ins VR128:$src1, RC:$src2),
469 !strconcat(base_opc, asm_opr),
470 [(set VR128:$dst, (vt (OpNode VR128:$src1,
471 (scalar_to_vector RC:$src2))))],
472 IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
474 // For the disassembler
475 let isCodeGenOnly = 1, hasSideEffects = 0 in
476 def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
477 (ins VR128:$src1, RC:$src2),
478 !strconcat(base_opc, asm_opr),
479 [], IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
482 multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt,
483 X86MemOperand x86memop, string OpcodeStr> {
485 defm V#NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
486 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">,
489 def V#NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
490 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
491 [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
492 VEX, VEX_LIG, Sched<[WriteStore]>;
494 let Constraints = "$src1 = $dst" in {
495 defm NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
496 "\t{$src2, $dst|$dst, $src2}">;
499 def NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
500 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
501 [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
505 // Loading from memory automatically zeroing upper bits.
506 multiclass sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
507 PatFrag mem_pat, string OpcodeStr> {
508 def V#NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
509 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
510 [(set RC:$dst, (mem_pat addr:$src))],
511 IIC_SSE_MOV_S_RM>, VEX, VEX_LIG, Sched<[WriteLoad]>;
512 def NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
513 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
514 [(set RC:$dst, (mem_pat addr:$src))],
515 IIC_SSE_MOV_S_RM>, Sched<[WriteLoad]>;
518 defm MOVSS : sse12_move<FR32, X86Movss, v4f32, f32mem, "movss">, XS;
519 defm MOVSD : sse12_move<FR64, X86Movsd, v2f64, f64mem, "movsd">, XD;
521 let canFoldAsLoad = 1, isReMaterializable = 1 in {
522 defm MOVSS : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
524 let AddedComplexity = 20 in
525 defm MOVSD : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
529 let Predicates = [UseAVX] in {
530 let AddedComplexity = 15 in {
531 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
532 // MOVS{S,D} to the lower bits.
533 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
534 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
535 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
536 (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
537 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
538 (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
539 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
540 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
542 // Move low f32 and clear high bits.
543 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
544 (SUBREG_TO_REG (i32 0),
545 (VMOVSSrr (v4f32 (V_SET0)),
546 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
547 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
548 (SUBREG_TO_REG (i32 0),
549 (VMOVSSrr (v4i32 (V_SET0)),
550 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
553 let AddedComplexity = 20 in {
554 // MOVSSrm zeros the high parts of the register; represent this
555 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
556 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
557 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
558 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
559 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
560 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
561 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
563 // MOVSDrm zeros the high parts of the register; represent this
564 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
565 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
566 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
567 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
568 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
569 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
570 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
571 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
572 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
573 def : Pat<(v2f64 (X86vzload addr:$src)),
574 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
576 // Represent the same patterns above but in the form they appear for
578 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
579 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (iPTR 0)))),
580 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
581 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
582 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (iPTR 0)))),
583 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
584 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
585 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))),
586 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
588 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
589 (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
590 (SUBREG_TO_REG (i32 0),
591 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
593 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
594 (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
595 (SUBREG_TO_REG (i64 0),
596 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
598 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
599 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
600 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
602 // Move low f64 and clear high bits.
603 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
604 (SUBREG_TO_REG (i32 0),
605 (VMOVSDrr (v2f64 (V_SET0)),
606 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
608 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
609 (SUBREG_TO_REG (i32 0),
610 (VMOVSDrr (v2i64 (V_SET0)),
611 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
613 // Extract and store.
614 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
616 (VMOVSSmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32))>;
617 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
619 (VMOVSDmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64))>;
621 // Shuffle with VMOVSS
622 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
623 (VMOVSSrr (v4i32 VR128:$src1),
624 (COPY_TO_REGCLASS (v4i32 VR128:$src2), FR32))>;
625 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
626 (VMOVSSrr (v4f32 VR128:$src1),
627 (COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
630 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
631 (SUBREG_TO_REG (i32 0),
632 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
633 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
635 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
636 (SUBREG_TO_REG (i32 0),
637 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
638 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
641 // Shuffle with VMOVSD
642 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
643 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
644 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
645 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
646 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
647 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
648 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
649 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
652 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
653 (SUBREG_TO_REG (i32 0),
654 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
655 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
657 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
658 (SUBREG_TO_REG (i32 0),
659 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
660 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
664 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
665 // is during lowering, where it's not possible to recognize the fold cause
666 // it has two uses through a bitcast. One use disappears at isel time and the
667 // fold opportunity reappears.
668 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
669 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
670 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
671 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
672 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
673 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
674 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
675 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
678 let Predicates = [UseSSE1] in {
679 let AddedComplexity = 15 in {
680 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
681 // MOVSS to the lower bits.
682 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
683 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
684 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
685 (MOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
686 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
687 (MOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
690 let AddedComplexity = 20 in {
691 // MOVSSrm already zeros the high parts of the register.
692 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
693 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
694 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
695 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
696 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
697 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
700 // Extract and store.
701 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
703 (MOVSSmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR32))>;
705 // Shuffle with MOVSS
706 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
707 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
708 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
709 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
712 let Predicates = [UseSSE2] in {
713 let AddedComplexity = 15 in {
714 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
715 // MOVSD to the lower bits.
716 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
717 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
720 let AddedComplexity = 20 in {
721 // MOVSDrm already zeros the high parts of the register.
722 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
723 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
724 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
725 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
726 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
727 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
728 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
729 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
730 def : Pat<(v2f64 (X86vzload addr:$src)),
731 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
734 // Extract and store.
735 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
737 (MOVSDmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR64))>;
739 // Shuffle with MOVSD
740 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
741 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
742 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
743 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
744 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
745 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
746 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
747 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
749 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
750 // is during lowering, where it's not possible to recognize the fold cause
751 // it has two uses through a bitcast. One use disappears at isel time and the
752 // fold opportunity reappears.
753 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
754 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
755 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
756 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
757 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
758 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
759 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
760 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
763 //===----------------------------------------------------------------------===//
764 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
765 //===----------------------------------------------------------------------===//
767 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
768 X86MemOperand x86memop, PatFrag ld_frag,
769 string asm, Domain d,
771 bit IsReMaterializable = 1> {
772 let neverHasSideEffects = 1 in
773 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
774 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>,
776 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
777 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
778 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
779 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>,
783 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
784 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
786 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
787 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
789 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
790 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
792 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
793 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
796 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
797 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
799 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
800 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
801 TB, OpSize, VEX, VEX_L;
802 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
803 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
805 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
806 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
807 TB, OpSize, VEX, VEX_L;
808 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
809 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
811 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
812 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
814 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
815 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
817 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
818 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
821 let SchedRW = [WriteStore] in {
822 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
823 "movaps\t{$src, $dst|$dst, $src}",
824 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
825 IIC_SSE_MOVA_P_MR>, VEX;
826 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
827 "movapd\t{$src, $dst|$dst, $src}",
828 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
829 IIC_SSE_MOVA_P_MR>, VEX;
830 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
831 "movups\t{$src, $dst|$dst, $src}",
832 [(store (v4f32 VR128:$src), addr:$dst)],
833 IIC_SSE_MOVU_P_MR>, VEX;
834 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
835 "movupd\t{$src, $dst|$dst, $src}",
836 [(store (v2f64 VR128:$src), addr:$dst)],
837 IIC_SSE_MOVU_P_MR>, VEX;
838 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
839 "movaps\t{$src, $dst|$dst, $src}",
840 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
841 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
842 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
843 "movapd\t{$src, $dst|$dst, $src}",
844 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
845 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
846 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
847 "movups\t{$src, $dst|$dst, $src}",
848 [(store (v8f32 VR256:$src), addr:$dst)],
849 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
850 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
851 "movupd\t{$src, $dst|$dst, $src}",
852 [(store (v4f64 VR256:$src), addr:$dst)],
853 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
857 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
858 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
860 "movaps\t{$src, $dst|$dst, $src}", [],
861 IIC_SSE_MOVA_P_RR>, VEX;
862 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
864 "movapd\t{$src, $dst|$dst, $src}", [],
865 IIC_SSE_MOVA_P_RR>, VEX;
866 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
868 "movups\t{$src, $dst|$dst, $src}", [],
869 IIC_SSE_MOVU_P_RR>, VEX;
870 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
872 "movupd\t{$src, $dst|$dst, $src}", [],
873 IIC_SSE_MOVU_P_RR>, VEX;
874 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
876 "movaps\t{$src, $dst|$dst, $src}", [],
877 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
878 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
880 "movapd\t{$src, $dst|$dst, $src}", [],
881 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
882 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
884 "movups\t{$src, $dst|$dst, $src}", [],
885 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
886 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
888 "movupd\t{$src, $dst|$dst, $src}", [],
889 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
892 let Predicates = [HasAVX] in {
893 def : Pat<(v8i32 (X86vzmovl
894 (insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)))),
895 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
896 def : Pat<(v4i64 (X86vzmovl
897 (insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)))),
898 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
899 def : Pat<(v8f32 (X86vzmovl
900 (insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)))),
901 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
902 def : Pat<(v4f64 (X86vzmovl
903 (insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)))),
904 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
908 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
909 (VMOVUPSYmr addr:$dst, VR256:$src)>;
910 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
911 (VMOVUPDYmr addr:$dst, VR256:$src)>;
913 let SchedRW = [WriteStore] in {
914 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
915 "movaps\t{$src, $dst|$dst, $src}",
916 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
918 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
919 "movapd\t{$src, $dst|$dst, $src}",
920 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
922 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
923 "movups\t{$src, $dst|$dst, $src}",
924 [(store (v4f32 VR128:$src), addr:$dst)],
926 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
927 "movupd\t{$src, $dst|$dst, $src}",
928 [(store (v2f64 VR128:$src), addr:$dst)],
933 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
934 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
935 "movaps\t{$src, $dst|$dst, $src}", [],
937 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
938 "movapd\t{$src, $dst|$dst, $src}", [],
940 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
941 "movups\t{$src, $dst|$dst, $src}", [],
943 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
944 "movupd\t{$src, $dst|$dst, $src}", [],
948 let Predicates = [HasAVX] in {
949 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
950 (VMOVUPSmr addr:$dst, VR128:$src)>;
951 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
952 (VMOVUPDmr addr:$dst, VR128:$src)>;
955 let Predicates = [UseSSE1] in
956 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
957 (MOVUPSmr addr:$dst, VR128:$src)>;
958 let Predicates = [UseSSE2] in
959 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
960 (MOVUPDmr addr:$dst, VR128:$src)>;
962 // Use vmovaps/vmovups for AVX integer load/store.
963 let Predicates = [HasAVX] in {
964 // 128-bit load/store
965 def : Pat<(alignedloadv2i64 addr:$src),
966 (VMOVAPSrm addr:$src)>;
967 def : Pat<(loadv2i64 addr:$src),
968 (VMOVUPSrm addr:$src)>;
970 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
971 (VMOVAPSmr addr:$dst, VR128:$src)>;
972 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
973 (VMOVAPSmr addr:$dst, VR128:$src)>;
974 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
975 (VMOVAPSmr addr:$dst, VR128:$src)>;
976 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
977 (VMOVAPSmr addr:$dst, VR128:$src)>;
978 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
979 (VMOVUPSmr addr:$dst, VR128:$src)>;
980 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
981 (VMOVUPSmr addr:$dst, VR128:$src)>;
982 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
983 (VMOVUPSmr addr:$dst, VR128:$src)>;
984 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
985 (VMOVUPSmr addr:$dst, VR128:$src)>;
987 // 256-bit load/store
988 def : Pat<(alignedloadv4i64 addr:$src),
989 (VMOVAPSYrm addr:$src)>;
990 def : Pat<(loadv4i64 addr:$src),
991 (VMOVUPSYrm addr:$src)>;
992 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
993 (VMOVAPSYmr addr:$dst, VR256:$src)>;
994 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
995 (VMOVAPSYmr addr:$dst, VR256:$src)>;
996 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
997 (VMOVAPSYmr addr:$dst, VR256:$src)>;
998 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
999 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1000 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1001 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1002 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1003 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1004 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1005 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1006 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1007 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1009 // Special patterns for storing subvector extracts of lower 128-bits
1010 // Its cheaper to just use VMOVAPS/VMOVUPS instead of VEXTRACTF128mr
1011 def : Pat<(alignedstore (v2f64 (extract_subvector
1012 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1013 (VMOVAPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1014 def : Pat<(alignedstore (v4f32 (extract_subvector
1015 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1016 (VMOVAPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1017 def : Pat<(alignedstore (v2i64 (extract_subvector
1018 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1019 (VMOVAPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1020 def : Pat<(alignedstore (v4i32 (extract_subvector
1021 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1022 (VMOVAPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1023 def : Pat<(alignedstore (v8i16 (extract_subvector
1024 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1025 (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1026 def : Pat<(alignedstore (v16i8 (extract_subvector
1027 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1028 (VMOVAPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1030 def : Pat<(store (v2f64 (extract_subvector
1031 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1032 (VMOVUPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1033 def : Pat<(store (v4f32 (extract_subvector
1034 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1035 (VMOVUPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1036 def : Pat<(store (v2i64 (extract_subvector
1037 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1038 (VMOVUPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1039 def : Pat<(store (v4i32 (extract_subvector
1040 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1041 (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1042 def : Pat<(store (v8i16 (extract_subvector
1043 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1044 (VMOVUPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1045 def : Pat<(store (v16i8 (extract_subvector
1046 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1047 (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1050 // Use movaps / movups for SSE integer load / store (one byte shorter).
1051 // The instructions selected below are then converted to MOVDQA/MOVDQU
1052 // during the SSE domain pass.
1053 let Predicates = [UseSSE1] in {
1054 def : Pat<(alignedloadv2i64 addr:$src),
1055 (MOVAPSrm addr:$src)>;
1056 def : Pat<(loadv2i64 addr:$src),
1057 (MOVUPSrm addr:$src)>;
1059 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1060 (MOVAPSmr addr:$dst, VR128:$src)>;
1061 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1062 (MOVAPSmr addr:$dst, VR128:$src)>;
1063 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1064 (MOVAPSmr addr:$dst, VR128:$src)>;
1065 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1066 (MOVAPSmr addr:$dst, VR128:$src)>;
1067 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1068 (MOVUPSmr addr:$dst, VR128:$src)>;
1069 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1070 (MOVUPSmr addr:$dst, VR128:$src)>;
1071 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1072 (MOVUPSmr addr:$dst, VR128:$src)>;
1073 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1074 (MOVUPSmr addr:$dst, VR128:$src)>;
1077 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1078 // bits are disregarded. FIXME: Set encoding to pseudo!
1079 let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
1080 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1081 "movaps\t{$src, $dst|$dst, $src}", [],
1082 IIC_SSE_MOVA_P_RR>, VEX;
1083 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1084 "movapd\t{$src, $dst|$dst, $src}", [],
1085 IIC_SSE_MOVA_P_RR>, VEX;
1086 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1087 "movaps\t{$src, $dst|$dst, $src}", [],
1089 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1090 "movapd\t{$src, $dst|$dst, $src}", [],
1094 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1095 // bits are disregarded. FIXME: Set encoding to pseudo!
1096 let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in {
1097 let isCodeGenOnly = 1 in {
1098 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1099 "movaps\t{$src, $dst|$dst, $src}",
1100 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1101 IIC_SSE_MOVA_P_RM>, VEX;
1102 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1103 "movapd\t{$src, $dst|$dst, $src}",
1104 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1105 IIC_SSE_MOVA_P_RM>, VEX;
1107 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1108 "movaps\t{$src, $dst|$dst, $src}",
1109 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1111 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1112 "movapd\t{$src, $dst|$dst, $src}",
1113 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1117 //===----------------------------------------------------------------------===//
1118 // SSE 1 & 2 - Move Low packed FP Instructions
1119 //===----------------------------------------------------------------------===//
1121 multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
1122 string base_opc, string asm_opr,
1123 InstrItinClass itin> {
1124 def PSrm : PI<opc, MRMSrcMem,
1125 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1126 !strconcat(base_opc, "s", asm_opr),
1128 (psnode VR128:$src1,
1129 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1130 itin, SSEPackedSingle>, TB,
1131 Sched<[WriteShuffleLd, ReadAfterLd]>;
1133 def PDrm : PI<opc, MRMSrcMem,
1134 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1135 !strconcat(base_opc, "d", asm_opr),
1136 [(set VR128:$dst, (v2f64 (pdnode VR128:$src1,
1137 (scalar_to_vector (loadf64 addr:$src2)))))],
1138 itin, SSEPackedDouble>, TB, OpSize,
1139 Sched<[WriteShuffleLd, ReadAfterLd]>;
1143 multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
1144 string base_opc, InstrItinClass itin> {
1145 defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1146 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1149 let Constraints = "$src1 = $dst" in
1150 defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1151 "\t{$src2, $dst|$dst, $src2}",
1155 let AddedComplexity = 20 in {
1156 defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp",
1160 let SchedRW = [WriteStore] in {
1161 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1162 "movlps\t{$src, $dst|$dst, $src}",
1163 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1164 (iPTR 0))), addr:$dst)],
1165 IIC_SSE_MOV_LH>, VEX;
1166 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1167 "movlpd\t{$src, $dst|$dst, $src}",
1168 [(store (f64 (vector_extract (v2f64 VR128:$src),
1169 (iPTR 0))), addr:$dst)],
1170 IIC_SSE_MOV_LH>, VEX;
1171 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1172 "movlps\t{$src, $dst|$dst, $src}",
1173 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1174 (iPTR 0))), addr:$dst)],
1176 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1177 "movlpd\t{$src, $dst|$dst, $src}",
1178 [(store (f64 (vector_extract (v2f64 VR128:$src),
1179 (iPTR 0))), addr:$dst)],
1183 let Predicates = [HasAVX] in {
1184 // Shuffle with VMOVLPS
1185 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1186 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1187 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1188 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1190 // Shuffle with VMOVLPD
1191 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1192 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1193 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1194 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1197 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1199 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1200 def : Pat<(store (v4i32 (X86Movlps
1201 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1202 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1203 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1205 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1206 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1208 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1211 let Predicates = [UseSSE1] in {
1212 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1213 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1214 (iPTR 0))), addr:$src1),
1215 (MOVLPSmr addr:$src1, VR128:$src2)>;
1217 // Shuffle with MOVLPS
1218 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1219 (MOVLPSrm VR128:$src1, addr:$src2)>;
1220 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1221 (MOVLPSrm VR128:$src1, addr:$src2)>;
1222 def : Pat<(X86Movlps VR128:$src1,
1223 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1224 (MOVLPSrm VR128:$src1, addr:$src2)>;
1227 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1229 (MOVLPSmr addr:$src1, VR128:$src2)>;
1230 def : Pat<(store (v4i32 (X86Movlps
1231 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1233 (MOVLPSmr addr:$src1, VR128:$src2)>;
1236 let Predicates = [UseSSE2] in {
1237 // Shuffle with MOVLPD
1238 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1239 (MOVLPDrm VR128:$src1, addr:$src2)>;
1240 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1241 (MOVLPDrm VR128:$src1, addr:$src2)>;
1244 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1246 (MOVLPDmr addr:$src1, VR128:$src2)>;
1247 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1249 (MOVLPDmr addr:$src1, VR128:$src2)>;
1252 //===----------------------------------------------------------------------===//
1253 // SSE 1 & 2 - Move Hi packed FP Instructions
1254 //===----------------------------------------------------------------------===//
1256 let AddedComplexity = 20 in {
1257 defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp",
1261 let SchedRW = [WriteStore] in {
1262 // v2f64 extract element 1 is always custom lowered to unpack high to low
1263 // and extract element 0 so the non-store version isn't too horrible.
1264 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1265 "movhps\t{$src, $dst|$dst, $src}",
1266 [(store (f64 (vector_extract
1267 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1268 (bc_v2f64 (v4f32 VR128:$src))),
1269 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1270 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1271 "movhpd\t{$src, $dst|$dst, $src}",
1272 [(store (f64 (vector_extract
1273 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1274 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1275 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1276 "movhps\t{$src, $dst|$dst, $src}",
1277 [(store (f64 (vector_extract
1278 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1279 (bc_v2f64 (v4f32 VR128:$src))),
1280 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1281 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1282 "movhpd\t{$src, $dst|$dst, $src}",
1283 [(store (f64 (vector_extract
1284 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1285 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1288 let Predicates = [HasAVX] in {
1290 def : Pat<(X86Movlhps VR128:$src1,
1291 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1292 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1293 def : Pat<(X86Movlhps VR128:$src1,
1294 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1295 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1297 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1298 // is during lowering, where it's not possible to recognize the load fold
1299 // cause it has two uses through a bitcast. One use disappears at isel time
1300 // and the fold opportunity reappears.
1301 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1302 (scalar_to_vector (loadf64 addr:$src2)))),
1303 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1306 let Predicates = [UseSSE1] in {
1308 def : Pat<(X86Movlhps VR128:$src1,
1309 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1310 (MOVHPSrm VR128:$src1, addr:$src2)>;
1311 def : Pat<(X86Movlhps VR128:$src1,
1312 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1313 (MOVHPSrm VR128:$src1, addr:$src2)>;
1316 let Predicates = [UseSSE2] in {
1317 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1318 // is during lowering, where it's not possible to recognize the load fold
1319 // cause it has two uses through a bitcast. One use disappears at isel time
1320 // and the fold opportunity reappears.
1321 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1322 (scalar_to_vector (loadf64 addr:$src2)))),
1323 (MOVHPDrm VR128:$src1, addr:$src2)>;
1326 //===----------------------------------------------------------------------===//
1327 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1328 //===----------------------------------------------------------------------===//
1330 let AddedComplexity = 20, Predicates = [UseAVX] in {
1331 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1332 (ins VR128:$src1, VR128:$src2),
1333 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1335 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1337 VEX_4V, Sched<[WriteShuffle]>;
1338 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1339 (ins VR128:$src1, VR128:$src2),
1340 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1342 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1344 VEX_4V, Sched<[WriteShuffle]>;
1346 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1347 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1348 (ins VR128:$src1, VR128:$src2),
1349 "movlhps\t{$src2, $dst|$dst, $src2}",
1351 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1352 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
1353 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1354 (ins VR128:$src1, VR128:$src2),
1355 "movhlps\t{$src2, $dst|$dst, $src2}",
1357 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1358 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
1361 let Predicates = [UseAVX] in {
1363 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1364 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1365 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1366 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1369 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1370 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1373 let Predicates = [UseSSE1] in {
1375 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1376 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1377 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1378 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1381 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1382 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1385 //===----------------------------------------------------------------------===//
1386 // SSE 1 & 2 - Conversion Instructions
1387 //===----------------------------------------------------------------------===//
1389 def SSE_CVT_PD : OpndItins<
1390 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1393 let Sched = WriteCvtI2F in
1394 def SSE_CVT_PS : OpndItins<
1395 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1398 let Sched = WriteCvtI2F in
1399 def SSE_CVT_Scalar : OpndItins<
1400 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1403 let Sched = WriteCvtF2I in
1404 def SSE_CVT_SS2SI_32 : OpndItins<
1405 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1408 let Sched = WriteCvtF2I in
1409 def SSE_CVT_SS2SI_64 : OpndItins<
1410 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1413 let Sched = WriteCvtF2I in
1414 def SSE_CVT_SD2SI : OpndItins<
1415 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1418 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1419 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1420 string asm, OpndItins itins> {
1421 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1422 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1423 itins.rr>, Sched<[itins.Sched]>;
1424 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1425 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1426 itins.rm>, Sched<[itins.Sched.Folded]>;
1429 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1430 X86MemOperand x86memop, string asm, Domain d,
1432 let neverHasSideEffects = 1 in {
1433 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1434 [], itins.rr, d>, Sched<[itins.Sched]>;
1436 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1437 [], itins.rm, d>, Sched<[itins.Sched.Folded]>;
1441 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1442 X86MemOperand x86memop, string asm> {
1443 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1444 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1445 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
1446 Sched<[WriteCvtI2F]>;
1448 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1449 (ins DstRC:$src1, x86memop:$src),
1450 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
1451 Sched<[WriteCvtI2FLd, ReadAfterLd]>;
1452 } // neverHasSideEffects = 1
1455 let Predicates = [UseAVX] in {
1456 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1457 "cvttss2si\t{$src, $dst|$dst, $src}",
1460 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1461 "cvttss2si\t{$src, $dst|$dst, $src}",
1463 XS, VEX, VEX_W, VEX_LIG;
1464 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1465 "cvttsd2si\t{$src, $dst|$dst, $src}",
1468 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1469 "cvttsd2si\t{$src, $dst|$dst, $src}",
1471 XD, VEX, VEX_W, VEX_LIG;
1473 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1474 (VCVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1475 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1476 (VCVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1477 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1478 (VCVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1479 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1480 (VCVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1481 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1482 (VCVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1483 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1484 (VCVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1485 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1486 (VCVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1487 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1488 (VCVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1490 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1491 // register, but the same isn't true when only using memory operands,
1492 // provide other assembly "l" and "q" forms to address this explicitly
1493 // where appropriate to do so.
1494 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss{l}">,
1495 XS, VEX_4V, VEX_LIG;
1496 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1497 XS, VEX_4V, VEX_W, VEX_LIG;
1498 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
1499 XD, VEX_4V, VEX_LIG;
1500 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1501 XD, VEX_4V, VEX_W, VEX_LIG;
1503 let Predicates = [UseAVX] in {
1504 def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1505 (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1506 def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1507 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1509 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1510 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1511 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1512 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1513 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1514 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1515 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1516 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1518 def : Pat<(f32 (sint_to_fp GR32:$src)),
1519 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1520 def : Pat<(f32 (sint_to_fp GR64:$src)),
1521 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1522 def : Pat<(f64 (sint_to_fp GR32:$src)),
1523 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1524 def : Pat<(f64 (sint_to_fp GR64:$src)),
1525 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1528 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1529 "cvttss2si\t{$src, $dst|$dst, $src}",
1530 SSE_CVT_SS2SI_32>, XS;
1531 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1532 "cvttss2si\t{$src, $dst|$dst, $src}",
1533 SSE_CVT_SS2SI_64>, XS, REX_W;
1534 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1535 "cvttsd2si\t{$src, $dst|$dst, $src}",
1537 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1538 "cvttsd2si\t{$src, $dst|$dst, $src}",
1539 SSE_CVT_SD2SI>, XD, REX_W;
1540 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1541 "cvtsi2ss{l}\t{$src, $dst|$dst, $src}",
1542 SSE_CVT_Scalar>, XS;
1543 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1544 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1545 SSE_CVT_Scalar>, XS, REX_W;
1546 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1547 "cvtsi2sd{l}\t{$src, $dst|$dst, $src}",
1548 SSE_CVT_Scalar>, XD;
1549 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1550 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1551 SSE_CVT_Scalar>, XD, REX_W;
1553 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1554 (CVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1555 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1556 (CVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1557 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1558 (CVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1559 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1560 (CVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1561 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1562 (CVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1563 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1564 (CVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1565 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1566 (CVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1567 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1568 (CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1570 def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}",
1571 (CVTSI2SSrm FR64:$dst, i32mem:$src)>;
1572 def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}",
1573 (CVTSI2SDrm FR64:$dst, i32mem:$src)>;
1575 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1576 // and/or XMM operand(s).
1578 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1579 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1580 string asm, OpndItins itins> {
1581 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1582 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1583 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>,
1584 Sched<[itins.Sched]>;
1585 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1586 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1587 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>,
1588 Sched<[itins.Sched.Folded]>;
1591 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1592 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1593 PatFrag ld_frag, string asm, OpndItins itins,
1595 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1597 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1598 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1599 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1600 itins.rr>, Sched<[itins.Sched]>;
1601 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1602 (ins DstRC:$src1, x86memop:$src2),
1604 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1605 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1606 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1607 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
1610 let Predicates = [UseAVX] in {
1611 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1612 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si",
1613 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1614 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1615 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si",
1616 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1618 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1619 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD;
1620 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1621 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1624 let Predicates = [UseAVX] in {
1625 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1626 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}",
1627 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1628 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1629 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1630 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1632 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1633 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}",
1634 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1635 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1636 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1637 SSE_CVT_Scalar, 0>, XD,
1640 let Constraints = "$src1 = $dst" in {
1641 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1642 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1643 "cvtsi2ss{l}", SSE_CVT_Scalar>, XS;
1644 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1645 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1646 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1647 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1648 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1649 "cvtsi2sd{l}", SSE_CVT_Scalar>, XD;
1650 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1651 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1652 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1657 // Aliases for intrinsics
1658 let Predicates = [UseAVX] in {
1659 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1660 ssmem, sse_load_f32, "cvttss2si",
1661 SSE_CVT_SS2SI_32>, XS, VEX;
1662 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1663 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1664 "cvttss2si", SSE_CVT_SS2SI_64>,
1666 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1667 sdmem, sse_load_f64, "cvttsd2si",
1668 SSE_CVT_SD2SI>, XD, VEX;
1669 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1670 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1671 "cvttsd2si", SSE_CVT_SD2SI>,
1674 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1675 ssmem, sse_load_f32, "cvttss2si",
1676 SSE_CVT_SS2SI_32>, XS;
1677 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1678 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1679 "cvttss2si", SSE_CVT_SS2SI_64>, XS, REX_W;
1680 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1681 sdmem, sse_load_f64, "cvttsd2si",
1683 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1684 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1685 "cvttsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1687 let Predicates = [UseAVX] in {
1688 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1689 ssmem, sse_load_f32, "cvtss2si",
1690 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1691 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1692 ssmem, sse_load_f32, "cvtss2si",
1693 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1695 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1696 ssmem, sse_load_f32, "cvtss2si",
1697 SSE_CVT_SS2SI_32>, XS;
1698 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1699 ssmem, sse_load_f32, "cvtss2si",
1700 SSE_CVT_SS2SI_64>, XS, REX_W;
1702 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1703 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1704 SSEPackedSingle, SSE_CVT_PS>,
1705 TB, VEX, Requires<[HasAVX]>;
1706 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
1707 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1708 SSEPackedSingle, SSE_CVT_PS>,
1709 TB, VEX, VEX_L, Requires<[HasAVX]>;
1711 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1712 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1713 SSEPackedSingle, SSE_CVT_PS>,
1714 TB, Requires<[UseSSE2]>;
1716 let Predicates = [UseAVX] in {
1717 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1718 (VCVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1719 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1720 (VCVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1721 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1722 (VCVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1723 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1724 (VCVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1725 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1726 (VCVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1727 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1728 (VCVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1729 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1730 (VCVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1731 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1732 (VCVTSD2SI64rm GR64:$dst, sdmem:$src), 0>;
1735 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1736 (CVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1737 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1738 (CVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1739 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1740 (CVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1741 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1742 (CVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1743 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1744 (CVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1745 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1746 (CVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1747 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1748 (CVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1749 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1750 (CVTSD2SI64rm GR64:$dst, sdmem:$src)>;
1754 // Convert scalar double to scalar single
1755 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1756 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1757 (ins FR64:$src1, FR64:$src2),
1758 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1759 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG,
1760 Sched<[WriteCvtF2F]>;
1762 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1763 (ins FR64:$src1, f64mem:$src2),
1764 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1765 [], IIC_SSE_CVT_Scalar_RM>,
1766 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG,
1767 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1770 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1773 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1774 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1775 [(set FR32:$dst, (fround FR64:$src))],
1776 IIC_SSE_CVT_Scalar_RR>, Sched<[WriteCvtF2F]>;
1777 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1778 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1779 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1780 IIC_SSE_CVT_Scalar_RM>,
1782 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
1784 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
1785 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1786 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1788 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1789 IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[UseAVX]>,
1790 Sched<[WriteCvtF2F]>;
1791 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
1792 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1793 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1794 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1795 VR128:$src1, sse_load_f64:$src2))],
1796 IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[UseAVX]>,
1797 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1799 let Constraints = "$src1 = $dst" in {
1800 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
1801 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1802 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1804 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1805 IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>,
1806 Sched<[WriteCvtF2F]>;
1807 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
1808 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1809 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1810 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1811 VR128:$src1, sse_load_f64:$src2))],
1812 IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>,
1813 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1816 // Convert scalar single to scalar double
1817 // SSE2 instructions with XS prefix
1818 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1819 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1820 (ins FR32:$src1, FR32:$src2),
1821 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1822 [], IIC_SSE_CVT_Scalar_RR>,
1823 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG,
1824 Sched<[WriteCvtF2F]>;
1826 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1827 (ins FR32:$src1, f32mem:$src2),
1828 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1829 [], IIC_SSE_CVT_Scalar_RM>,
1830 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>,
1831 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1834 def : Pat<(f64 (fextend FR32:$src)),
1835 (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[UseAVX]>;
1836 def : Pat<(fextend (loadf32 addr:$src)),
1837 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[UseAVX]>;
1839 def : Pat<(extloadf32 addr:$src),
1840 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
1841 Requires<[UseAVX, OptForSize]>;
1842 def : Pat<(extloadf32 addr:$src),
1843 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
1844 Requires<[UseAVX, OptForSpeed]>;
1846 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1847 "cvtss2sd\t{$src, $dst|$dst, $src}",
1848 [(set FR64:$dst, (fextend FR32:$src))],
1849 IIC_SSE_CVT_Scalar_RR>, XS,
1850 Requires<[UseSSE2]>, Sched<[WriteCvtF2F]>;
1851 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1852 "cvtss2sd\t{$src, $dst|$dst, $src}",
1853 [(set FR64:$dst, (extloadf32 addr:$src))],
1854 IIC_SSE_CVT_Scalar_RM>, XS,
1855 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
1857 // extload f32 -> f64. This matches load+fextend because we have a hack in
1858 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1860 // Since these loads aren't folded into the fextend, we have to match it
1862 def : Pat<(fextend (loadf32 addr:$src)),
1863 (CVTSS2SDrm addr:$src)>, Requires<[UseSSE2]>;
1864 def : Pat<(extloadf32 addr:$src),
1865 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[UseSSE2, OptForSpeed]>;
1867 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1868 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1869 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1871 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1872 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[UseAVX]>,
1873 Sched<[WriteCvtF2F]>;
1874 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1875 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1876 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1878 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1879 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[UseAVX]>,
1880 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1881 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1882 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1883 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1884 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1886 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1887 IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>,
1888 Sched<[WriteCvtF2F]>;
1889 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1890 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1891 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1893 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1894 IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>,
1895 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1898 // Convert packed single/double fp to doubleword
1899 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1900 "cvtps2dq\t{$src, $dst|$dst, $src}",
1901 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1902 IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
1903 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1904 "cvtps2dq\t{$src, $dst|$dst, $src}",
1906 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1907 IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
1908 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1909 "cvtps2dq\t{$src, $dst|$dst, $src}",
1911 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1912 IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
1913 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1914 "cvtps2dq\t{$src, $dst|$dst, $src}",
1916 (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
1917 IIC_SSE_CVT_PS_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
1918 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1919 "cvtps2dq\t{$src, $dst|$dst, $src}",
1920 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1921 IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
1922 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1923 "cvtps2dq\t{$src, $dst|$dst, $src}",
1925 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1926 IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
1929 // Convert Packed Double FP to Packed DW Integers
1930 let Predicates = [HasAVX] in {
1931 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1932 // register, but the same isn't true when using memory operands instead.
1933 // Provide other assembly rr and rm forms to address this explicitly.
1934 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1935 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1936 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1937 VEX, Sched<[WriteCvtF2I]>;
1940 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1941 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1942 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1943 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1945 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX,
1946 Sched<[WriteCvtF2ILd]>;
1949 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1950 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1952 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L,
1953 Sched<[WriteCvtF2I]>;
1954 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1955 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1957 (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
1958 VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
1959 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1960 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1963 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1964 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1966 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1967 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2ILd]>;
1968 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1969 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1970 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1971 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
1973 // Convert with truncation packed single/double fp to doubleword
1974 // SSE2 packed instructions with XS prefix
1975 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1976 "cvttps2dq\t{$src, $dst|$dst, $src}",
1978 (int_x86_sse2_cvttps2dq VR128:$src))],
1979 IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
1980 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1981 "cvttps2dq\t{$src, $dst|$dst, $src}",
1982 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1983 (memopv4f32 addr:$src)))],
1984 IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
1985 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1986 "cvttps2dq\t{$src, $dst|$dst, $src}",
1988 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1989 IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
1990 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1991 "cvttps2dq\t{$src, $dst|$dst, $src}",
1992 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1993 (memopv8f32 addr:$src)))],
1994 IIC_SSE_CVT_PS_RM>, VEX, VEX_L,
1995 Sched<[WriteCvtF2ILd]>;
1997 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1998 "cvttps2dq\t{$src, $dst|$dst, $src}",
1999 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
2000 IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
2001 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2002 "cvttps2dq\t{$src, $dst|$dst, $src}",
2004 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
2005 IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
2007 let Predicates = [HasAVX] in {
2008 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
2009 (VCVTDQ2PSrr VR128:$src)>;
2010 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2011 (VCVTDQ2PSrm addr:$src)>;
2013 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
2014 (VCVTDQ2PSrr VR128:$src)>;
2015 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
2016 (VCVTDQ2PSrm addr:$src)>;
2018 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2019 (VCVTTPS2DQrr VR128:$src)>;
2020 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
2021 (VCVTTPS2DQrm addr:$src)>;
2023 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
2024 (VCVTDQ2PSYrr VR256:$src)>;
2025 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
2026 (VCVTDQ2PSYrm addr:$src)>;
2028 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
2029 (VCVTTPS2DQYrr VR256:$src)>;
2030 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
2031 (VCVTTPS2DQYrm addr:$src)>;
2034 let Predicates = [UseSSE2] in {
2035 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
2036 (CVTDQ2PSrr VR128:$src)>;
2037 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2038 (CVTDQ2PSrm addr:$src)>;
2040 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
2041 (CVTDQ2PSrr VR128:$src)>;
2042 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
2043 (CVTDQ2PSrm addr:$src)>;
2045 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2046 (CVTTPS2DQrr VR128:$src)>;
2047 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
2048 (CVTTPS2DQrm addr:$src)>;
2051 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2052 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2054 (int_x86_sse2_cvttpd2dq VR128:$src))],
2055 IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2I]>;
2057 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2058 // register, but the same isn't true when using memory operands instead.
2059 // Provide other assembly rr and rm forms to address this explicitly.
2062 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
2063 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
2064 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2065 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
2066 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2067 (memopv2f64 addr:$src)))],
2068 IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2ILd]>;
2071 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2072 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2074 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
2075 IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
2076 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2077 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2079 (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
2080 IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
2081 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
2082 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
2084 let Predicates = [HasAVX] in {
2085 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
2086 (VCVTTPD2DQYrr VR256:$src)>;
2087 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
2088 (VCVTTPD2DQYrm addr:$src)>;
2089 } // Predicates = [HasAVX]
2091 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2092 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2093 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
2094 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
2095 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
2096 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2097 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2098 (memopv2f64 addr:$src)))],
2100 Sched<[WriteCvtF2ILd]>;
2102 // Convert packed single to packed double
2103 let Predicates = [HasAVX] in {
2104 // SSE2 instructions without OpSize prefix
2105 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2106 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2107 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2108 IIC_SSE_CVT_PD_RR>, TB, VEX, Sched<[WriteCvtF2F]>;
2109 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2110 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2111 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2112 IIC_SSE_CVT_PD_RM>, TB, VEX, Sched<[WriteCvtF2FLd]>;
2113 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2114 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2116 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
2117 IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L, Sched<[WriteCvtF2F]>;
2118 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
2119 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2121 (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
2122 IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
2125 let Predicates = [UseSSE2] in {
2126 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2127 "cvtps2pd\t{$src, $dst|$dst, $src}",
2128 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2129 IIC_SSE_CVT_PD_RR>, TB, Sched<[WriteCvtF2F]>;
2130 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2131 "cvtps2pd\t{$src, $dst|$dst, $src}",
2132 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2133 IIC_SSE_CVT_PD_RM>, TB, Sched<[WriteCvtF2FLd]>;
2136 // Convert Packed DW Integers to Packed Double FP
2137 let Predicates = [HasAVX] in {
2138 let neverHasSideEffects = 1, mayLoad = 1 in
2139 def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2140 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2141 []>, VEX, Sched<[WriteCvtI2FLd]>;
2142 def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2143 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2145 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX,
2146 Sched<[WriteCvtI2F]>;
2147 def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2148 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2150 (int_x86_avx_cvtdq2_pd_256
2151 (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L,
2152 Sched<[WriteCvtI2FLd]>;
2153 def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2154 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2156 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L,
2157 Sched<[WriteCvtI2F]>;
2160 let neverHasSideEffects = 1, mayLoad = 1 in
2161 def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2162 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2163 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtI2FLd]>;
2164 def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2165 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2166 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2167 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtI2F]>;
2169 // AVX 256-bit register conversion intrinsics
2170 let Predicates = [HasAVX] in {
2171 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2172 (VCVTDQ2PDYrr VR128:$src)>;
2173 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2174 (VCVTDQ2PDYrm addr:$src)>;
2175 } // Predicates = [HasAVX]
2177 // Convert packed double to packed single
2178 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2179 // register, but the same isn't true when using memory operands instead.
2180 // Provide other assembly rr and rm forms to address this explicitly.
2181 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2182 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2183 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2184 IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2F]>;
2187 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2188 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2189 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2190 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2192 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2193 IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2FLd]>;
2196 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2197 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2199 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2200 IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2F]>;
2201 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2202 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2204 (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
2205 IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
2206 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2207 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2209 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2210 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2211 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2212 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2F]>;
2213 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2214 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2216 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2217 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2FLd]>;
2220 // AVX 256-bit register conversion intrinsics
2221 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2222 // whenever possible to avoid declaring two versions of each one.
2223 let Predicates = [HasAVX] in {
2224 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2225 (VCVTDQ2PSYrr VR256:$src)>;
2226 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2227 (VCVTDQ2PSYrm addr:$src)>;
2229 // Match fround and fextend for 128/256-bit conversions
2230 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2231 (VCVTPD2PSrr VR128:$src)>;
2232 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2233 (VCVTPD2PSXrm addr:$src)>;
2234 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2235 (VCVTPD2PSYrr VR256:$src)>;
2236 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2237 (VCVTPD2PSYrm addr:$src)>;
2239 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2240 (VCVTPS2PDrr VR128:$src)>;
2241 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2242 (VCVTPS2PDYrr VR128:$src)>;
2243 def : Pat<(v4f64 (extloadv4f32 addr:$src)),
2244 (VCVTPS2PDYrm addr:$src)>;
2247 let Predicates = [UseSSE2] in {
2248 // Match fround and fextend for 128 conversions
2249 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2250 (CVTPD2PSrr VR128:$src)>;
2251 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2252 (CVTPD2PSrm addr:$src)>;
2254 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2255 (CVTPS2PDrr VR128:$src)>;
2258 //===----------------------------------------------------------------------===//
2259 // SSE 1 & 2 - Compare Instructions
2260 //===----------------------------------------------------------------------===//
2262 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2263 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2264 Operand CC, SDNode OpNode, ValueType VT,
2265 PatFrag ld_frag, string asm, string asm_alt,
2267 def rr : SIi8<0xC2, MRMSrcReg,
2268 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2269 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2270 itins.rr>, Sched<[itins.Sched]>;
2271 def rm : SIi8<0xC2, MRMSrcMem,
2272 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2273 [(set RC:$dst, (OpNode (VT RC:$src1),
2274 (ld_frag addr:$src2), imm:$cc))],
2276 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2278 // Accept explicit immediate argument form instead of comparison code.
2279 let neverHasSideEffects = 1 in {
2280 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2281 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2282 IIC_SSE_ALU_F32S_RR>, Sched<[itins.Sched]>;
2284 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2285 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2286 IIC_SSE_ALU_F32S_RM>,
2287 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2291 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2292 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2293 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2295 XS, VEX_4V, VEX_LIG;
2296 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2297 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2298 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2299 SSE_ALU_F32S>, // same latency as 32 bit compare
2300 XD, VEX_4V, VEX_LIG;
2302 let Constraints = "$src1 = $dst" in {
2303 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2304 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2305 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2307 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2308 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2309 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2310 SSE_ALU_F32S>, // same latency as 32 bit compare
2314 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2315 Intrinsic Int, string asm, OpndItins itins> {
2316 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2317 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2318 [(set VR128:$dst, (Int VR128:$src1,
2319 VR128:$src, imm:$cc))],
2321 Sched<[itins.Sched]>;
2322 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2323 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2324 [(set VR128:$dst, (Int VR128:$src1,
2325 (load addr:$src), imm:$cc))],
2327 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2330 // Aliases to match intrinsics which expect XMM operand(s).
2331 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2332 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2335 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2336 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2337 SSE_ALU_F32S>, // same latency as f32
2339 let Constraints = "$src1 = $dst" in {
2340 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2341 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2343 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2344 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2345 SSE_ALU_F32S>, // same latency as f32
2350 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2351 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2352 ValueType vt, X86MemOperand x86memop,
2353 PatFrag ld_frag, string OpcodeStr> {
2354 def rr: SI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2355 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2356 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2359 def rm: SI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2360 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2361 [(set EFLAGS, (OpNode (vt RC:$src1),
2362 (ld_frag addr:$src2)))],
2364 Sched<[WriteFAddLd, ReadAfterLd]>;
2367 let Defs = [EFLAGS] in {
2368 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2369 "ucomiss">, TB, VEX, VEX_LIG;
2370 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2371 "ucomisd">, TB, OpSize, VEX, VEX_LIG;
2372 let Pattern = []<dag> in {
2373 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2374 "comiss">, TB, VEX, VEX_LIG;
2375 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2376 "comisd">, TB, OpSize, VEX, VEX_LIG;
2379 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2380 load, "ucomiss">, TB, VEX;
2381 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2382 load, "ucomisd">, TB, OpSize, VEX;
2384 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2385 load, "comiss">, TB, VEX;
2386 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2387 load, "comisd">, TB, OpSize, VEX;
2388 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2390 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2391 "ucomisd">, TB, OpSize;
2393 let Pattern = []<dag> in {
2394 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2396 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2397 "comisd">, TB, OpSize;
2400 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2401 load, "ucomiss">, TB;
2402 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2403 load, "ucomisd">, TB, OpSize;
2405 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2407 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2408 "comisd">, TB, OpSize;
2409 } // Defs = [EFLAGS]
2411 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2412 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2413 Operand CC, Intrinsic Int, string asm,
2414 string asm_alt, Domain d> {
2415 def rri : PIi8<0xC2, MRMSrcReg,
2416 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2417 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2418 IIC_SSE_CMPP_RR, d>,
2420 def rmi : PIi8<0xC2, MRMSrcMem,
2421 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2422 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2423 IIC_SSE_CMPP_RM, d>,
2424 Sched<[WriteFAddLd, ReadAfterLd]>;
2426 // Accept explicit immediate argument form instead of comparison code.
2427 let neverHasSideEffects = 1 in {
2428 def rri_alt : PIi8<0xC2, MRMSrcReg,
2429 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2430 asm_alt, [], IIC_SSE_CMPP_RR, d>, Sched<[WriteFAdd]>;
2431 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2432 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2433 asm_alt, [], IIC_SSE_CMPP_RM, d>,
2434 Sched<[WriteFAddLd, ReadAfterLd]>;
2438 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2439 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2440 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2441 SSEPackedSingle>, TB, VEX_4V;
2442 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2443 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2444 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2445 SSEPackedDouble>, TB, OpSize, VEX_4V;
2446 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2447 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2448 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2449 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2450 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2451 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2452 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2453 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2454 let Constraints = "$src1 = $dst" in {
2455 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2456 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2457 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2458 SSEPackedSingle>, TB;
2459 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2460 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2461 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2462 SSEPackedDouble>, TB, OpSize;
2465 let Predicates = [HasAVX] in {
2466 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2467 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2468 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2469 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2470 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2471 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2472 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2473 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2475 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2476 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2477 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2478 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2479 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2480 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2481 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2482 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2485 let Predicates = [UseSSE1] in {
2486 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2487 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2488 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2489 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2492 let Predicates = [UseSSE2] in {
2493 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2494 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2495 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2496 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2499 //===----------------------------------------------------------------------===//
2500 // SSE 1 & 2 - Shuffle Instructions
2501 //===----------------------------------------------------------------------===//
2503 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2504 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2505 ValueType vt, string asm, PatFrag mem_frag,
2506 Domain d, bit IsConvertibleToThreeAddress = 0> {
2507 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2508 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2509 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2510 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
2511 Sched<[WriteShuffleLd, ReadAfterLd]>;
2512 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2513 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2514 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2515 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2516 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
2517 Sched<[WriteShuffle]>;
2520 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2521 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2522 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2523 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2524 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2525 memopv8f32, SSEPackedSingle>, TB, VEX_4V, VEX_L;
2526 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2527 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2528 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2529 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2530 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2531 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2533 let Constraints = "$src1 = $dst" in {
2534 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2535 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2536 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2538 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2539 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2540 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2544 let Predicates = [HasAVX] in {
2545 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2546 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2547 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2548 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2549 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2551 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2552 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2553 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2554 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2555 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2558 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2559 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2560 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2561 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2562 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2564 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2565 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2566 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2567 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2568 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2571 let Predicates = [UseSSE1] in {
2572 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2573 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2574 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2575 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2576 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2579 let Predicates = [UseSSE2] in {
2580 // Generic SHUFPD patterns
2581 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2582 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2583 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2584 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2585 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2588 //===----------------------------------------------------------------------===//
2589 // SSE 1 & 2 - Unpack Instructions
2590 //===----------------------------------------------------------------------===//
2592 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2593 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2594 PatFrag mem_frag, RegisterClass RC,
2595 X86MemOperand x86memop, string asm,
2597 def rr : PI<opc, MRMSrcReg,
2598 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2600 (vt (OpNode RC:$src1, RC:$src2)))],
2601 IIC_SSE_UNPCK, d>, Sched<[WriteShuffle]>;
2602 def rm : PI<opc, MRMSrcMem,
2603 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2605 (vt (OpNode RC:$src1,
2606 (mem_frag addr:$src2))))],
2608 Sched<[WriteShuffleLd, ReadAfterLd]>;
2611 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2612 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2613 SSEPackedSingle>, TB, VEX_4V;
2614 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2615 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2616 SSEPackedDouble>, TB, OpSize, VEX_4V;
2617 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2618 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2619 SSEPackedSingle>, TB, VEX_4V;
2620 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2621 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2622 SSEPackedDouble>, TB, OpSize, VEX_4V;
2624 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2625 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2626 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2627 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2628 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2629 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2630 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2631 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2632 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2633 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2634 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2635 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2637 let Constraints = "$src1 = $dst" in {
2638 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2639 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2640 SSEPackedSingle>, TB;
2641 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2642 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2643 SSEPackedDouble>, TB, OpSize;
2644 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2645 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2646 SSEPackedSingle>, TB;
2647 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2648 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2649 SSEPackedDouble>, TB, OpSize;
2650 } // Constraints = "$src1 = $dst"
2652 let Predicates = [HasAVX1Only] in {
2653 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2654 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2655 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
2656 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2657 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2658 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2659 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
2660 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2662 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
2663 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2664 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
2665 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2666 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
2667 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2668 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
2669 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2672 let Predicates = [HasAVX] in {
2673 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2674 // problem is during lowering, where it's not possible to recognize the load
2675 // fold cause it has two uses through a bitcast. One use disappears at isel
2676 // time and the fold opportunity reappears.
2677 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2678 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2681 let Predicates = [UseSSE2] in {
2682 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2683 // problem is during lowering, where it's not possible to recognize the load
2684 // fold cause it has two uses through a bitcast. One use disappears at isel
2685 // time and the fold opportunity reappears.
2686 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2687 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2690 //===----------------------------------------------------------------------===//
2691 // SSE 1 & 2 - Extract Floating-Point Sign mask
2692 //===----------------------------------------------------------------------===//
2694 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2695 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2697 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2698 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2699 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>,
2700 Sched<[WriteVecLogic]>;
2701 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2702 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2703 IIC_SSE_MOVMSK, d>, REX_W, Sched<[WriteVecLogic]>;
2706 let Predicates = [HasAVX] in {
2707 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2708 "movmskps", SSEPackedSingle>, TB, VEX;
2709 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2710 "movmskpd", SSEPackedDouble>, TB,
2712 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2713 "movmskps", SSEPackedSingle>, TB,
2715 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2716 "movmskpd", SSEPackedDouble>, TB,
2719 def : Pat<(i32 (X86fgetsign FR32:$src)),
2720 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2721 def : Pat<(i64 (X86fgetsign FR32:$src)),
2722 (VMOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2723 def : Pat<(i32 (X86fgetsign FR64:$src)),
2724 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2725 def : Pat<(i64 (X86fgetsign FR64:$src)),
2726 (VMOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2729 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2730 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2731 SSEPackedSingle>, TB, VEX, Sched<[WriteVecLogic]>;
2732 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2733 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2734 SSEPackedDouble>, TB,
2735 OpSize, VEX, Sched<[WriteVecLogic]>;
2736 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2737 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2738 SSEPackedSingle>, TB, VEX, VEX_L, Sched<[WriteVecLogic]>;
2739 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2740 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2741 SSEPackedDouble>, TB,
2742 OpSize, VEX, VEX_L, Sched<[WriteVecLogic]>;
2745 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2746 SSEPackedSingle>, TB;
2747 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2748 SSEPackedDouble>, TB, OpSize;
2750 def : Pat<(i32 (X86fgetsign FR32:$src)),
2751 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2752 Requires<[UseSSE1]>;
2753 def : Pat<(i64 (X86fgetsign FR32:$src)),
2754 (MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2755 Requires<[UseSSE1]>;
2756 def : Pat<(i32 (X86fgetsign FR64:$src)),
2757 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2758 Requires<[UseSSE2]>;
2759 def : Pat<(i64 (X86fgetsign FR64:$src)),
2760 (MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2761 Requires<[UseSSE2]>;
2763 //===---------------------------------------------------------------------===//
2764 // SSE2 - Packed Integer Logical Instructions
2765 //===---------------------------------------------------------------------===//
2767 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2769 /// PDI_binop_rm - Simple SSE2 binary operator.
2770 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2771 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2772 X86MemOperand x86memop, OpndItins itins,
2773 bit IsCommutable, bit Is2Addr> {
2774 let isCommutable = IsCommutable in
2775 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2776 (ins RC:$src1, RC:$src2),
2778 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2779 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2780 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
2781 Sched<[itins.Sched]>;
2782 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2783 (ins RC:$src1, x86memop:$src2),
2785 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2786 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2787 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2788 (bitconvert (memop_frag addr:$src2)))))],
2790 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2792 } // ExeDomain = SSEPackedInt
2794 multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
2795 ValueType OpVT128, ValueType OpVT256,
2796 OpndItins itins, bit IsCommutable = 0> {
2797 let Predicates = [HasAVX] in
2798 defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
2799 VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
2801 let Constraints = "$src1 = $dst" in
2802 defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
2803 memopv2i64, i128mem, itins, IsCommutable, 1>;
2805 let Predicates = [HasAVX2] in
2806 defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
2807 OpVT256, VR256, memopv4i64, i256mem, itins,
2808 IsCommutable, 0>, VEX_4V, VEX_L;
2811 // These are ordered here for pattern ordering requirements with the fp versions
2813 defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2814 defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2815 defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2816 defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
2817 SSE_BIT_ITINS_P, 0>;
2819 //===----------------------------------------------------------------------===//
2820 // SSE 1 & 2 - Logical Instructions
2821 //===----------------------------------------------------------------------===//
2823 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2825 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2826 SDNode OpNode, OpndItins itins> {
2827 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2828 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2831 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2832 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2835 let Constraints = "$src1 = $dst" in {
2836 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2837 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2840 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2841 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2846 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2847 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2849 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2851 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2854 let isCommutable = 0 in
2855 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", X86fandn,
2858 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2860 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2862 defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2863 !strconcat(OpcodeStr, "ps"), f256mem,
2864 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2865 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2866 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
2868 defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2869 !strconcat(OpcodeStr, "pd"), f256mem,
2870 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2871 (bc_v4i64 (v4f64 VR256:$src2))))],
2872 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2873 (memopv4i64 addr:$src2)))], 0>,
2874 TB, OpSize, VEX_4V, VEX_L;
2876 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2877 // are all promoted to v2i64, and the patterns are covered by the int
2878 // version. This is needed in SSE only, because v2i64 isn't supported on
2879 // SSE1, but only on SSE2.
2880 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2881 !strconcat(OpcodeStr, "ps"), f128mem, [],
2882 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2883 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2885 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2886 !strconcat(OpcodeStr, "pd"), f128mem,
2887 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2888 (bc_v2i64 (v2f64 VR128:$src2))))],
2889 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2890 (memopv2i64 addr:$src2)))], 0>,
2893 let Constraints = "$src1 = $dst" in {
2894 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2895 !strconcat(OpcodeStr, "ps"), f128mem,
2896 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2897 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2898 (memopv2i64 addr:$src2)))]>, TB;
2900 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2901 !strconcat(OpcodeStr, "pd"), f128mem,
2902 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2903 (bc_v2i64 (v2f64 VR128:$src2))))],
2904 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2905 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2909 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2910 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2911 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2912 let isCommutable = 0 in
2913 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2915 //===----------------------------------------------------------------------===//
2916 // SSE 1 & 2 - Arithmetic Instructions
2917 //===----------------------------------------------------------------------===//
2919 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2922 /// In addition, we also have a special variant of the scalar form here to
2923 /// represent the associated intrinsic operation. This form is unlike the
2924 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2925 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2927 /// These three forms can each be reg+reg or reg+mem.
2930 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2932 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
2933 SDNode OpNode, SizeItins itins> {
2934 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2935 VR128, v4f32, f128mem, memopv4f32,
2936 SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
2937 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2938 VR128, v2f64, f128mem, memopv2f64,
2939 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V;
2941 defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
2942 OpNode, VR256, v8f32, f256mem, memopv8f32,
2943 SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
2944 defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
2945 OpNode, VR256, v4f64, f256mem, memopv4f64,
2946 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L;
2948 let Constraints = "$src1 = $dst" in {
2949 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2950 v4f32, f128mem, memopv4f32, SSEPackedSingle,
2952 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2953 v2f64, f128mem, memopv2f64, SSEPackedDouble,
2954 itins.d>, TB, OpSize;
2958 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2960 defm V#NAME#SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2961 OpNode, FR32, f32mem, itins.s, 0>, XS, VEX_4V, VEX_LIG;
2962 defm V#NAME#SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2963 OpNode, FR64, f64mem, itins.d, 0>, XD, VEX_4V, VEX_LIG;
2965 let Constraints = "$src1 = $dst" in {
2966 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2967 OpNode, FR32, f32mem, itins.s>, XS;
2968 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2969 OpNode, FR64, f64mem, itins.d>, XD;
2973 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2975 defm V#NAME#SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2976 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2977 itins.s, 0>, XS, VEX_4V, VEX_LIG;
2978 defm V#NAME#SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2979 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2980 itins.d, 0>, XD, VEX_4V, VEX_LIG;
2982 let Constraints = "$src1 = $dst" in {
2983 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2984 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2986 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2987 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2992 // Binary Arithmetic instructions
2993 defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2994 basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2995 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2996 defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2997 basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2998 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2999 let isCommutable = 0 in {
3000 defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
3001 basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
3002 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
3003 defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
3004 basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
3005 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
3006 defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
3007 basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
3008 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
3009 defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
3010 basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
3011 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
3014 let isCodeGenOnly = 1 in {
3015 defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>,
3016 basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
3017 defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>,
3018 basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
3022 /// In addition, we also have a special variant of the scalar form here to
3023 /// represent the associated intrinsic operation. This form is unlike the
3024 /// plain scalar form, in that it takes an entire vector (instead of a
3025 /// scalar) and leaves the top elements undefined.
3027 /// And, we have a special variant form for a full-vector intrinsic form.
3029 let Sched = WriteFSqrt in {
3030 def SSE_SQRTPS : OpndItins<
3031 IIC_SSE_SQRTPS_RR, IIC_SSE_SQRTPS_RM
3034 def SSE_SQRTSS : OpndItins<
3035 IIC_SSE_SQRTSS_RR, IIC_SSE_SQRTSS_RM
3038 def SSE_SQRTPD : OpndItins<
3039 IIC_SSE_SQRTPD_RR, IIC_SSE_SQRTPD_RM
3042 def SSE_SQRTSD : OpndItins<
3043 IIC_SSE_SQRTSD_RR, IIC_SSE_SQRTSD_RM
3047 let Sched = WriteFRcp in {
3048 def SSE_RCPP : OpndItins<
3049 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
3052 def SSE_RCPS : OpndItins<
3053 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
3057 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3058 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3059 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3060 let Predicates = [HasAVX], hasSideEffects = 0 in {
3061 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3062 (ins FR32:$src1, FR32:$src2),
3063 !strconcat("v", OpcodeStr,
3064 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3065 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3066 let mayLoad = 1 in {
3067 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3068 (ins FR32:$src1,f32mem:$src2),
3069 !strconcat("v", OpcodeStr,
3070 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3071 []>, VEX_4V, VEX_LIG,
3072 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3073 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3074 (ins VR128:$src1, ssmem:$src2),
3075 !strconcat("v", OpcodeStr,
3076 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3077 []>, VEX_4V, VEX_LIG,
3078 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3082 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3083 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3084 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3085 // For scalar unary operations, fold a load into the operation
3086 // only in OptForSize mode. It eliminates an instruction, but it also
3087 // eliminates a whole-register clobber (the load), so it introduces a
3088 // partial register update condition.
3089 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3090 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3091 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3092 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3093 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3094 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3095 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>,
3096 Sched<[itins.Sched]>;
3097 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3098 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3099 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>,
3100 Sched<[itins.Sched.Folded]>;
3103 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3104 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3106 let Predicates = [HasAVX], hasSideEffects = 0 in {
3107 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3108 (ins FR32:$src1, FR32:$src2),
3109 !strconcat("v", OpcodeStr,
3110 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3111 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3112 let mayLoad = 1 in {
3113 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3114 (ins FR32:$src1,f32mem:$src2),
3115 !strconcat("v", OpcodeStr,
3116 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3117 []>, VEX_4V, VEX_LIG,
3118 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3119 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3120 (ins VR128:$src1, ssmem:$src2),
3121 !strconcat("v", OpcodeStr,
3122 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3123 []>, VEX_4V, VEX_LIG,
3124 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3128 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3129 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3130 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3131 // For scalar unary operations, fold a load into the operation
3132 // only in OptForSize mode. It eliminates an instruction, but it also
3133 // eliminates a whole-register clobber (the load), so it introduces a
3134 // partial register update condition.
3135 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3136 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3137 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3138 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3139 let Constraints = "$src1 = $dst" in {
3140 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3141 (ins VR128:$src1, VR128:$src2),
3142 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3143 [], itins.rr>, Sched<[itins.Sched]>;
3144 let mayLoad = 1, hasSideEffects = 0 in
3145 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3146 (ins VR128:$src1, ssmem:$src2),
3147 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3148 [], itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3152 /// sse1_fp_unop_p - SSE1 unops in packed form.
3153 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3155 let Predicates = [HasAVX] in {
3156 def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3157 !strconcat("v", OpcodeStr,
3158 "ps\t{$src, $dst|$dst, $src}"),
3159 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
3160 itins.rr>, VEX, Sched<[itins.Sched]>;
3161 def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3162 !strconcat("v", OpcodeStr,
3163 "ps\t{$src, $dst|$dst, $src}"),
3164 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))],
3165 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3166 def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3167 !strconcat("v", OpcodeStr,
3168 "ps\t{$src, $dst|$dst, $src}"),
3169 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3170 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3171 def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3172 !strconcat("v", OpcodeStr,
3173 "ps\t{$src, $dst|$dst, $src}"),
3174 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3175 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3178 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3179 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3180 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>,
3181 Sched<[itins.Sched]>;
3182 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3183 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3184 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>,
3185 Sched<[itins.Sched.Folded]>;
3188 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3189 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3190 Intrinsic V4F32Int, Intrinsic V8F32Int,
3192 let Predicates = [HasAVX] in {
3193 def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3194 !strconcat("v", OpcodeStr,
3195 "ps\t{$src, $dst|$dst, $src}"),
3196 [(set VR128:$dst, (V4F32Int VR128:$src))],
3197 itins.rr>, VEX, Sched<[itins.Sched]>;
3198 def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3199 !strconcat("v", OpcodeStr,
3200 "ps\t{$src, $dst|$dst, $src}"),
3201 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3202 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3203 def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3204 !strconcat("v", OpcodeStr,
3205 "ps\t{$src, $dst|$dst, $src}"),
3206 [(set VR256:$dst, (V8F32Int VR256:$src))],
3207 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3208 def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
3210 !strconcat("v", OpcodeStr,
3211 "ps\t{$src, $dst|$dst, $src}"),
3212 [(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
3213 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3216 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3217 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3218 [(set VR128:$dst, (V4F32Int VR128:$src))],
3219 itins.rr>, Sched<[itins.Sched]>;
3220 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3221 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3222 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3223 itins.rm>, Sched<[itins.Sched.Folded]>;
3226 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3227 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3228 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3229 let Predicates = [HasAVX], hasSideEffects = 0 in {
3230 def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
3231 (ins FR64:$src1, FR64:$src2),
3232 !strconcat("v", OpcodeStr,
3233 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3234 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3235 let mayLoad = 1 in {
3236 def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
3237 (ins FR64:$src1,f64mem:$src2),
3238 !strconcat("v", OpcodeStr,
3239 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3240 []>, VEX_4V, VEX_LIG,
3241 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3242 def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3243 (ins VR128:$src1, sdmem:$src2),
3244 !strconcat("v", OpcodeStr,
3245 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3246 []>, VEX_4V, VEX_LIG,
3247 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3251 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3252 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3253 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>,
3254 Sched<[itins.Sched]>;
3255 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3256 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3257 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3258 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3259 Requires<[UseSSE2, OptForSize]>, Sched<[itins.Sched.Folded]>;
3260 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3261 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3262 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>,
3263 Sched<[itins.Sched]>;
3264 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3265 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3266 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>,
3267 Sched<[itins.Sched.Folded]>;
3270 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3271 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3272 SDNode OpNode, OpndItins itins> {
3273 let Predicates = [HasAVX] in {
3274 def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3275 !strconcat("v", OpcodeStr,
3276 "pd\t{$src, $dst|$dst, $src}"),
3277 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
3278 itins.rr>, VEX, Sched<[itins.Sched]>;
3279 def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3280 !strconcat("v", OpcodeStr,
3281 "pd\t{$src, $dst|$dst, $src}"),
3282 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))],
3283 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3284 def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3285 !strconcat("v", OpcodeStr,
3286 "pd\t{$src, $dst|$dst, $src}"),
3287 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3288 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3289 def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3290 !strconcat("v", OpcodeStr,
3291 "pd\t{$src, $dst|$dst, $src}"),
3292 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3293 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3296 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3297 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3298 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>,
3299 Sched<[itins.Sched]>;
3300 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3301 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3302 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>,
3303 Sched<[itins.Sched.Folded]>;
3307 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3309 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPS>,
3310 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3312 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPD>;
3314 // Reciprocal approximations. Note that these typically require refinement
3315 // in order to obtain suitable precision.
3316 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTSS>,
3317 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTPS>,
3318 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3319 int_x86_avx_rsqrt_ps_256, SSE_SQRTPS>;
3320 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
3321 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
3322 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
3323 int_x86_avx_rcp_ps_256, SSE_RCPP>;
3325 let Predicates = [UseAVX] in {
3326 def : Pat<(f32 (fsqrt FR32:$src)),
3327 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3328 def : Pat<(f32 (fsqrt (load addr:$src))),
3329 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3330 Requires<[HasAVX, OptForSize]>;
3331 def : Pat<(f64 (fsqrt FR64:$src)),
3332 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3333 def : Pat<(f64 (fsqrt (load addr:$src))),
3334 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3335 Requires<[HasAVX, OptForSize]>;
3337 def : Pat<(f32 (X86frsqrt FR32:$src)),
3338 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3339 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3340 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3341 Requires<[HasAVX, OptForSize]>;
3343 def : Pat<(f32 (X86frcp FR32:$src)),
3344 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3345 def : Pat<(f32 (X86frcp (load addr:$src))),
3346 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3347 Requires<[HasAVX, OptForSize]>;
3349 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3350 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3351 (COPY_TO_REGCLASS VR128:$src, FR32)),
3353 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3354 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3356 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3357 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3358 (COPY_TO_REGCLASS VR128:$src, FR64)),
3360 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3361 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3363 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3364 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3365 (COPY_TO_REGCLASS VR128:$src, FR32)),
3367 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3368 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3370 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3371 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3372 (COPY_TO_REGCLASS VR128:$src, FR32)),
3374 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3375 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3378 // Reciprocal approximations. Note that these typically require refinement
3379 // in order to obtain suitable precision.
3380 let Predicates = [UseSSE1] in {
3381 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3382 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3383 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3384 (RCPSSr_Int VR128:$src, VR128:$src)>;
3387 // There is no f64 version of the reciprocal approximation instructions.
3389 //===----------------------------------------------------------------------===//
3390 // SSE 1 & 2 - Non-temporal stores
3391 //===----------------------------------------------------------------------===//
3393 let AddedComplexity = 400 in { // Prefer non-temporal versions
3394 let SchedRW = [WriteStore] in {
3395 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3396 (ins f128mem:$dst, VR128:$src),
3397 "movntps\t{$src, $dst|$dst, $src}",
3398 [(alignednontemporalstore (v4f32 VR128:$src),
3400 IIC_SSE_MOVNT>, VEX;
3401 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3402 (ins f128mem:$dst, VR128:$src),
3403 "movntpd\t{$src, $dst|$dst, $src}",
3404 [(alignednontemporalstore (v2f64 VR128:$src),
3406 IIC_SSE_MOVNT>, VEX;
3408 let ExeDomain = SSEPackedInt in
3409 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3410 (ins f128mem:$dst, VR128:$src),
3411 "movntdq\t{$src, $dst|$dst, $src}",
3412 [(alignednontemporalstore (v2i64 VR128:$src),
3414 IIC_SSE_MOVNT>, VEX;
3416 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3417 (ins f256mem:$dst, VR256:$src),
3418 "movntps\t{$src, $dst|$dst, $src}",
3419 [(alignednontemporalstore (v8f32 VR256:$src),
3421 IIC_SSE_MOVNT>, VEX, VEX_L;
3422 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3423 (ins f256mem:$dst, VR256:$src),
3424 "movntpd\t{$src, $dst|$dst, $src}",
3425 [(alignednontemporalstore (v4f64 VR256:$src),
3427 IIC_SSE_MOVNT>, VEX, VEX_L;
3428 let ExeDomain = SSEPackedInt in
3429 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3430 (ins f256mem:$dst, VR256:$src),
3431 "movntdq\t{$src, $dst|$dst, $src}",
3432 [(alignednontemporalstore (v4i64 VR256:$src),
3434 IIC_SSE_MOVNT>, VEX, VEX_L;
3436 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3437 "movntps\t{$src, $dst|$dst, $src}",
3438 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3440 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3441 "movntpd\t{$src, $dst|$dst, $src}",
3442 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3445 let ExeDomain = SSEPackedInt in
3446 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3447 "movntdq\t{$src, $dst|$dst, $src}",
3448 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3451 // There is no AVX form for instructions below this point
3452 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3453 "movnti{l}\t{$src, $dst|$dst, $src}",
3454 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3456 TB, Requires<[HasSSE2]>;
3457 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3458 "movnti{q}\t{$src, $dst|$dst, $src}",
3459 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3461 TB, Requires<[HasSSE2]>;
3462 } // SchedRW = [WriteStore]
3464 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3465 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3467 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3468 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3469 } // AddedComplexity
3471 //===----------------------------------------------------------------------===//
3472 // SSE 1 & 2 - Prefetch and memory fence
3473 //===----------------------------------------------------------------------===//
3475 // Prefetch intrinsic.
3476 let Predicates = [HasSSE1], SchedRW = [WriteLoad] in {
3477 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3478 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3479 IIC_SSE_PREFETCH>, TB;
3480 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3481 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3482 IIC_SSE_PREFETCH>, TB;
3483 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3484 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3485 IIC_SSE_PREFETCH>, TB;
3486 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3487 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3488 IIC_SSE_PREFETCH>, TB;
3491 // FIXME: How should these memory instructions be modeled?
3492 let SchedRW = [WriteLoad] in {
3494 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3495 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3496 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3498 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3499 // was introduced with SSE2, it's backward compatible.
3500 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3502 // Load, store, and memory fence
3503 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3504 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3505 TB, Requires<[HasSSE1]>;
3506 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3507 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3508 TB, Requires<[HasSSE2]>;
3509 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3510 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3511 TB, Requires<[HasSSE2]>;
3514 def : Pat<(X86SFence), (SFENCE)>;
3515 def : Pat<(X86LFence), (LFENCE)>;
3516 def : Pat<(X86MFence), (MFENCE)>;
3518 //===----------------------------------------------------------------------===//
3519 // SSE 1 & 2 - Load/Store XCSR register
3520 //===----------------------------------------------------------------------===//
3522 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3523 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3524 IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>;
3525 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3526 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3527 IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>;
3529 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3530 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3531 IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>;
3532 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3533 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3534 IIC_SSE_STMXCSR>, Sched<[WriteStore]>;
3536 //===---------------------------------------------------------------------===//
3537 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3538 //===---------------------------------------------------------------------===//
3540 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3542 let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
3543 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3544 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3546 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3547 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3549 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3550 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3552 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3553 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3558 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
3559 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3560 "movdqa\t{$src, $dst|$dst, $src}", [],
3563 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3564 "movdqa\t{$src, $dst|$dst, $src}", [],
3565 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3566 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3567 "movdqu\t{$src, $dst|$dst, $src}", [],
3570 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3571 "movdqu\t{$src, $dst|$dst, $src}", [],
3572 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3575 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3576 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3577 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3578 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3580 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3581 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3583 let Predicates = [HasAVX] in {
3584 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3585 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3587 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3588 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3593 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3594 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3595 (ins i128mem:$dst, VR128:$src),
3596 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3598 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3599 (ins i256mem:$dst, VR256:$src),
3600 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3602 let Predicates = [HasAVX] in {
3603 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3604 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3606 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3607 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3612 let SchedRW = [WriteMove] in {
3613 let neverHasSideEffects = 1 in
3614 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3615 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3617 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3618 "movdqu\t{$src, $dst|$dst, $src}",
3619 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3622 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3623 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3624 "movdqa\t{$src, $dst|$dst, $src}", [],
3627 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3628 "movdqu\t{$src, $dst|$dst, $src}",
3629 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3633 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3634 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3635 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3636 "movdqa\t{$src, $dst|$dst, $src}",
3637 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3639 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3640 "movdqu\t{$src, $dst|$dst, $src}",
3641 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3643 XS, Requires<[UseSSE2]>;
3646 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3647 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3648 "movdqa\t{$src, $dst|$dst, $src}",
3649 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3651 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3652 "movdqu\t{$src, $dst|$dst, $src}",
3653 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3655 XS, Requires<[UseSSE2]>;
3658 } // ExeDomain = SSEPackedInt
3660 let Predicates = [HasAVX] in {
3661 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3662 (VMOVDQUmr addr:$dst, VR128:$src)>;
3663 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3664 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3666 let Predicates = [UseSSE2] in
3667 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3668 (MOVDQUmr addr:$dst, VR128:$src)>;
3670 //===---------------------------------------------------------------------===//
3671 // SSE2 - Packed Integer Arithmetic Instructions
3672 //===---------------------------------------------------------------------===//
3674 let Sched = WriteVecIMul in
3675 def SSE_PMADD : OpndItins<
3676 IIC_SSE_PMADD, IIC_SSE_PMADD
3679 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3681 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3682 RegisterClass RC, PatFrag memop_frag,
3683 X86MemOperand x86memop,
3685 bit IsCommutable = 0,
3687 let isCommutable = IsCommutable in
3688 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3689 (ins RC:$src1, RC:$src2),
3691 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3692 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3693 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>,
3694 Sched<[itins.Sched]>;
3695 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3696 (ins RC:$src1, x86memop:$src2),
3698 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3699 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3700 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3701 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3704 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3705 Intrinsic IntId256, OpndItins itins,
3706 bit IsCommutable = 0> {
3707 let Predicates = [HasAVX] in
3708 defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3709 VR128, memopv2i64, i128mem, itins,
3710 IsCommutable, 0>, VEX_4V;
3712 let Constraints = "$src1 = $dst" in
3713 defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3714 i128mem, itins, IsCommutable, 1>;
3716 let Predicates = [HasAVX2] in
3717 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3718 VR256, memopv4i64, i256mem, itins,
3719 IsCommutable, 0>, VEX_4V, VEX_L;
3722 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3723 string OpcodeStr, SDNode OpNode,
3724 SDNode OpNode2, RegisterClass RC,
3725 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3726 ShiftOpndItins itins,
3728 // src2 is always 128-bit
3729 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3730 (ins RC:$src1, VR128:$src2),
3732 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3733 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3734 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3735 itins.rr>, Sched<[WriteVecShift]>;
3736 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3737 (ins RC:$src1, i128mem:$src2),
3739 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3740 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3741 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3742 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>,
3743 Sched<[WriteVecShiftLd, ReadAfterLd]>;
3744 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3745 (ins RC:$src1, i32i8imm:$src2),
3747 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3748 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3749 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>,
3750 Sched<[WriteVecShift]>;
3753 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3754 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3755 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3756 PatFrag memop_frag, X86MemOperand x86memop,
3758 bit IsCommutable = 0, bit Is2Addr = 1> {
3759 let isCommutable = IsCommutable in
3760 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3761 (ins RC:$src1, RC:$src2),
3763 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3764 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3765 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
3766 Sched<[itins.Sched]>;
3767 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3768 (ins RC:$src1, x86memop:$src2),
3770 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3771 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3772 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3773 (bitconvert (memop_frag addr:$src2)))))]>,
3774 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3776 } // ExeDomain = SSEPackedInt
3778 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3779 SSE_INTALU_ITINS_P, 1>;
3780 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3781 SSE_INTALU_ITINS_P, 1>;
3782 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3783 SSE_INTALU_ITINS_P, 1>;
3784 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3785 SSE_INTALUQ_ITINS_P, 1>;
3786 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3787 SSE_INTMUL_ITINS_P, 1>;
3788 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3789 SSE_INTALU_ITINS_P, 0>;
3790 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3791 SSE_INTALU_ITINS_P, 0>;
3792 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3793 SSE_INTALU_ITINS_P, 0>;
3794 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3795 SSE_INTALUQ_ITINS_P, 0>;
3796 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3797 SSE_INTALU_ITINS_P, 0>;
3798 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3799 SSE_INTALU_ITINS_P, 0>;
3800 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3801 SSE_INTALU_ITINS_P, 1>;
3802 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3803 SSE_INTALU_ITINS_P, 1>;
3804 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3805 SSE_INTALU_ITINS_P, 1>;
3806 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3807 SSE_INTALU_ITINS_P, 1>;
3810 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3811 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3812 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3813 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3814 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3815 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3816 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3817 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3818 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3819 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3820 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3821 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3822 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3823 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3824 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3825 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3826 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3827 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3828 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3829 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3830 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3831 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3832 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3833 int_x86_avx2_psad_bw, SSE_INTALU_ITINS_P, 1>;
3835 let Predicates = [HasAVX] in
3836 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3837 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3839 let Predicates = [HasAVX2] in
3840 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3841 VR256, memopv4i64, i256mem,
3842 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3843 let Constraints = "$src1 = $dst" in
3844 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3845 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3847 //===---------------------------------------------------------------------===//
3848 // SSE2 - Packed Integer Logical Instructions
3849 //===---------------------------------------------------------------------===//
3851 let Predicates = [HasAVX] in {
3852 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3853 VR128, v8i16, v8i16, bc_v8i16,
3854 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3855 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3856 VR128, v4i32, v4i32, bc_v4i32,
3857 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3858 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3859 VR128, v2i64, v2i64, bc_v2i64,
3860 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3862 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3863 VR128, v8i16, v8i16, bc_v8i16,
3864 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3865 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3866 VR128, v4i32, v4i32, bc_v4i32,
3867 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3868 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3869 VR128, v2i64, v2i64, bc_v2i64,
3870 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3872 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3873 VR128, v8i16, v8i16, bc_v8i16,
3874 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3875 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3876 VR128, v4i32, v4i32, bc_v4i32,
3877 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3879 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3880 // 128-bit logical shifts.
3881 def VPSLLDQri : PDIi8<0x73, MRM7r,
3882 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3883 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3885 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3887 def VPSRLDQri : PDIi8<0x73, MRM3r,
3888 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3889 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3891 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3893 // PSRADQri doesn't exist in SSE[1-3].
3895 } // Predicates = [HasAVX]
3897 let Predicates = [HasAVX2] in {
3898 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3899 VR256, v16i16, v8i16, bc_v8i16,
3900 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3901 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3902 VR256, v8i32, v4i32, bc_v4i32,
3903 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3904 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3905 VR256, v4i64, v2i64, bc_v2i64,
3906 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3908 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3909 VR256, v16i16, v8i16, bc_v8i16,
3910 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3911 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3912 VR256, v8i32, v4i32, bc_v4i32,
3913 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3914 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3915 VR256, v4i64, v2i64, bc_v2i64,
3916 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3918 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3919 VR256, v16i16, v8i16, bc_v8i16,
3920 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3921 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3922 VR256, v8i32, v4i32, bc_v4i32,
3923 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3925 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3926 // 256-bit logical shifts.
3927 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3928 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3929 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3931 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3933 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3934 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3935 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3937 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3939 // PSRADQYri doesn't exist in SSE[1-3].
3941 } // Predicates = [HasAVX2]
3943 let Constraints = "$src1 = $dst" in {
3944 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3945 VR128, v8i16, v8i16, bc_v8i16,
3946 SSE_INTSHIFT_ITINS_P>;
3947 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3948 VR128, v4i32, v4i32, bc_v4i32,
3949 SSE_INTSHIFT_ITINS_P>;
3950 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3951 VR128, v2i64, v2i64, bc_v2i64,
3952 SSE_INTSHIFT_ITINS_P>;
3954 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3955 VR128, v8i16, v8i16, bc_v8i16,
3956 SSE_INTSHIFT_ITINS_P>;
3957 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3958 VR128, v4i32, v4i32, bc_v4i32,
3959 SSE_INTSHIFT_ITINS_P>;
3960 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3961 VR128, v2i64, v2i64, bc_v2i64,
3962 SSE_INTSHIFT_ITINS_P>;
3964 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3965 VR128, v8i16, v8i16, bc_v8i16,
3966 SSE_INTSHIFT_ITINS_P>;
3967 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3968 VR128, v4i32, v4i32, bc_v4i32,
3969 SSE_INTSHIFT_ITINS_P>;
3971 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3972 // 128-bit logical shifts.
3973 def PSLLDQri : PDIi8<0x73, MRM7r,
3974 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3975 "pslldq\t{$src2, $dst|$dst, $src2}",
3977 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3978 def PSRLDQri : PDIi8<0x73, MRM3r,
3979 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3980 "psrldq\t{$src2, $dst|$dst, $src2}",
3982 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3983 // PSRADQri doesn't exist in SSE[1-3].
3985 } // Constraints = "$src1 = $dst"
3987 let Predicates = [HasAVX] in {
3988 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3989 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3990 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3991 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3992 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3993 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3995 // Shift up / down and insert zero's.
3996 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3997 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3998 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3999 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4002 let Predicates = [HasAVX2] in {
4003 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4004 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4005 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4006 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4009 let Predicates = [UseSSE2] in {
4010 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4011 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4012 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4013 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4014 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4015 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4017 // Shift up / down and insert zero's.
4018 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4019 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4020 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4021 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4024 //===---------------------------------------------------------------------===//
4025 // SSE2 - Packed Integer Comparison Instructions
4026 //===---------------------------------------------------------------------===//
4028 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
4029 SSE_INTALU_ITINS_P, 1>;
4030 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
4031 SSE_INTALU_ITINS_P, 1>;
4032 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
4033 SSE_INTALU_ITINS_P, 1>;
4034 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
4035 SSE_INTALU_ITINS_P, 0>;
4036 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
4037 SSE_INTALU_ITINS_P, 0>;
4038 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
4039 SSE_INTALU_ITINS_P, 0>;
4041 //===---------------------------------------------------------------------===//
4042 // SSE2 - Packed Integer Pack Instructions
4043 //===---------------------------------------------------------------------===//
4045 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4046 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
4047 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4048 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
4049 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4050 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
4052 //===---------------------------------------------------------------------===//
4053 // SSE2 - Packed Integer Shuffle Instructions
4054 //===---------------------------------------------------------------------===//
4056 let ExeDomain = SSEPackedInt in {
4057 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
4059 let Predicates = [HasAVX] in {
4060 def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
4061 (ins VR128:$src1, i8imm:$src2),
4062 !strconcat("v", OpcodeStr,
4063 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4065 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4066 IIC_SSE_PSHUF>, VEX, Sched<[WriteShuffle]>;
4067 def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
4068 (ins i128mem:$src1, i8imm:$src2),
4069 !strconcat("v", OpcodeStr,
4070 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4072 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4073 (i8 imm:$src2))))], IIC_SSE_PSHUF>, VEX,
4074 Sched<[WriteShuffleLd]>;
4077 let Predicates = [HasAVX2] in {
4078 def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
4079 (ins VR256:$src1, i8imm:$src2),
4080 !strconcat("v", OpcodeStr,
4081 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4083 (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
4084 IIC_SSE_PSHUF>, VEX, VEX_L, Sched<[WriteShuffle]>;
4085 def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
4086 (ins i256mem:$src1, i8imm:$src2),
4087 !strconcat("v", OpcodeStr,
4088 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4090 (vt256 (OpNode (bitconvert (memopv4i64 addr:$src1)),
4091 (i8 imm:$src2))))], IIC_SSE_PSHUF>, VEX, VEX_L,
4092 Sched<[WriteShuffleLd]>;
4095 let Predicates = [UseSSE2] in {
4096 def ri : Ii8<0x70, MRMSrcReg,
4097 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4098 !strconcat(OpcodeStr,
4099 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4101 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4102 IIC_SSE_PSHUF>, Sched<[WriteShuffle]>;
4103 def mi : Ii8<0x70, MRMSrcMem,
4104 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4105 !strconcat(OpcodeStr,
4106 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4108 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4109 (i8 imm:$src2))))], IIC_SSE_PSHUF>,
4110 Sched<[WriteShuffleLd]>;
4113 } // ExeDomain = SSEPackedInt
4115 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, TB, OpSize;
4116 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS;
4117 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD;
4119 let Predicates = [HasAVX] in {
4120 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4121 (VPSHUFDmi addr:$src1, imm:$imm)>;
4122 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4123 (VPSHUFDri VR128:$src1, imm:$imm)>;
4126 let Predicates = [UseSSE2] in {
4127 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4128 (PSHUFDmi addr:$src1, imm:$imm)>;
4129 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4130 (PSHUFDri VR128:$src1, imm:$imm)>;
4133 //===---------------------------------------------------------------------===//
4134 // SSE2 - Packed Integer Unpack Instructions
4135 //===---------------------------------------------------------------------===//
4137 let ExeDomain = SSEPackedInt in {
4138 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4139 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4140 def rr : PDI<opc, MRMSrcReg,
4141 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4143 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4144 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4145 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4146 IIC_SSE_UNPCK>, Sched<[WriteShuffle]>;
4147 def rm : PDI<opc, MRMSrcMem,
4148 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4150 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4151 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4152 [(set VR128:$dst, (OpNode VR128:$src1,
4153 (bc_frag (memopv2i64
4156 Sched<[WriteShuffleLd, ReadAfterLd]>;
4159 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4160 SDNode OpNode, PatFrag bc_frag> {
4161 def Yrr : PDI<opc, MRMSrcReg,
4162 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4163 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4164 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>,
4165 Sched<[WriteShuffle]>;
4166 def Yrm : PDI<opc, MRMSrcMem,
4167 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4168 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4169 [(set VR256:$dst, (OpNode VR256:$src1,
4170 (bc_frag (memopv4i64 addr:$src2))))]>,
4171 Sched<[WriteShuffleLd, ReadAfterLd]>;
4174 let Predicates = [HasAVX] in {
4175 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4176 bc_v16i8, 0>, VEX_4V;
4177 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4178 bc_v8i16, 0>, VEX_4V;
4179 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4180 bc_v4i32, 0>, VEX_4V;
4181 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4182 bc_v2i64, 0>, VEX_4V;
4184 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4185 bc_v16i8, 0>, VEX_4V;
4186 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4187 bc_v8i16, 0>, VEX_4V;
4188 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4189 bc_v4i32, 0>, VEX_4V;
4190 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4191 bc_v2i64, 0>, VEX_4V;
4194 let Predicates = [HasAVX2] in {
4195 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4196 bc_v32i8>, VEX_4V, VEX_L;
4197 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4198 bc_v16i16>, VEX_4V, VEX_L;
4199 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4200 bc_v8i32>, VEX_4V, VEX_L;
4201 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4202 bc_v4i64>, VEX_4V, VEX_L;
4204 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4205 bc_v32i8>, VEX_4V, VEX_L;
4206 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4207 bc_v16i16>, VEX_4V, VEX_L;
4208 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4209 bc_v8i32>, VEX_4V, VEX_L;
4210 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4211 bc_v4i64>, VEX_4V, VEX_L;
4214 let Constraints = "$src1 = $dst" in {
4215 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4217 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4219 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4221 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4224 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4226 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4228 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4230 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4233 } // ExeDomain = SSEPackedInt
4235 //===---------------------------------------------------------------------===//
4236 // SSE2 - Packed Integer Extract and Insert
4237 //===---------------------------------------------------------------------===//
4239 let ExeDomain = SSEPackedInt in {
4240 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4241 def rri : Ii8<0xC4, MRMSrcReg,
4242 (outs VR128:$dst), (ins VR128:$src1,
4243 GR32:$src2, i32i8imm:$src3),
4245 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4246 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4248 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>,
4249 Sched<[WriteShuffle]>;
4250 def rmi : Ii8<0xC4, MRMSrcMem,
4251 (outs VR128:$dst), (ins VR128:$src1,
4252 i16mem:$src2, i32i8imm:$src3),
4254 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4255 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4257 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4258 imm:$src3))], IIC_SSE_PINSRW>,
4259 Sched<[WriteShuffleLd, ReadAfterLd]>;
4263 let Predicates = [HasAVX] in
4264 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4265 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4266 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4267 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4268 imm:$src2))]>, TB, OpSize, VEX,
4269 Sched<[WriteShuffle]>;
4270 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4271 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4272 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4273 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4274 imm:$src2))], IIC_SSE_PEXTRW>,
4275 Sched<[WriteShuffleLd, ReadAfterLd]>;
4278 let Predicates = [HasAVX] in {
4279 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4280 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4281 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4282 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4283 []>, TB, OpSize, VEX_4V, Sched<[WriteShuffle]>;
4286 let Constraints = "$src1 = $dst" in
4287 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[UseSSE2]>;
4289 } // ExeDomain = SSEPackedInt
4291 //===---------------------------------------------------------------------===//
4292 // SSE2 - Packed Mask Creation
4293 //===---------------------------------------------------------------------===//
4295 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecLogic] in {
4297 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4298 "pmovmskb\t{$src, $dst|$dst, $src}",
4299 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4300 IIC_SSE_MOVMSK>, VEX;
4301 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4302 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4304 let Predicates = [HasAVX2] in {
4305 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4306 "pmovmskb\t{$src, $dst|$dst, $src}",
4307 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX, VEX_L;
4308 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4309 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4312 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4313 "pmovmskb\t{$src, $dst|$dst, $src}",
4314 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4317 } // ExeDomain = SSEPackedInt
4319 //===---------------------------------------------------------------------===//
4320 // SSE2 - Conditional Store
4321 //===---------------------------------------------------------------------===//
4323 let ExeDomain = SSEPackedInt, SchedRW = [WriteStore] in {
4326 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4327 (ins VR128:$src, VR128:$mask),
4328 "maskmovdqu\t{$mask, $src|$src, $mask}",
4329 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4330 IIC_SSE_MASKMOV>, VEX;
4332 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4333 (ins VR128:$src, VR128:$mask),
4334 "maskmovdqu\t{$mask, $src|$src, $mask}",
4335 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4336 IIC_SSE_MASKMOV>, VEX;
4339 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4340 "maskmovdqu\t{$mask, $src|$src, $mask}",
4341 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4344 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4345 "maskmovdqu\t{$mask, $src|$src, $mask}",
4346 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4349 } // ExeDomain = SSEPackedInt
4351 //===---------------------------------------------------------------------===//
4352 // SSE2 - Move Doubleword
4353 //===---------------------------------------------------------------------===//
4355 //===---------------------------------------------------------------------===//
4356 // Move Int Doubleword to Packed Double Int
4358 def VMOVDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4359 "movd\t{$src, $dst|$dst, $src}",
4361 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4362 VEX, Sched<[WriteMove]>;
4363 def VMOVDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4364 "movd\t{$src, $dst|$dst, $src}",
4366 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4368 VEX, Sched<[WriteLoad]>;
4369 def VMOV64toPQIrr : VRS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4370 "mov{d|q}\t{$src, $dst|$dst, $src}",
4372 (v2i64 (scalar_to_vector GR64:$src)))],
4373 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4374 def VMOV64toSDrr : VRS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4375 "mov{d|q}\t{$src, $dst|$dst, $src}",
4376 [(set FR64:$dst, (bitconvert GR64:$src))],
4377 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4379 def MOVDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4380 "movd\t{$src, $dst|$dst, $src}",
4382 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4384 def MOVDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4385 "movd\t{$src, $dst|$dst, $src}",
4387 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4388 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4389 def MOV64toPQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4390 "mov{d|q}\t{$src, $dst|$dst, $src}",
4392 (v2i64 (scalar_to_vector GR64:$src)))],
4393 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4394 def MOV64toSDrr : RS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4395 "mov{d|q}\t{$src, $dst|$dst, $src}",
4396 [(set FR64:$dst, (bitconvert GR64:$src))],
4397 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4399 //===---------------------------------------------------------------------===//
4400 // Move Int Doubleword to Single Scalar
4402 def VMOVDI2SSrr : VS2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4403 "movd\t{$src, $dst|$dst, $src}",
4404 [(set FR32:$dst, (bitconvert GR32:$src))],
4405 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4407 def VMOVDI2SSrm : VS2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4408 "movd\t{$src, $dst|$dst, $src}",
4409 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4411 VEX, Sched<[WriteLoad]>;
4412 def MOVDI2SSrr : S2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4413 "movd\t{$src, $dst|$dst, $src}",
4414 [(set FR32:$dst, (bitconvert GR32:$src))],
4415 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4417 def MOVDI2SSrm : S2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4418 "movd\t{$src, $dst|$dst, $src}",
4419 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4420 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4422 //===---------------------------------------------------------------------===//
4423 // Move Packed Doubleword Int to Packed Double Int
4425 def VMOVPDI2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4426 "movd\t{$src, $dst|$dst, $src}",
4427 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4428 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX,
4430 def VMOVPDI2DImr : VS2I<0x7E, MRMDestMem, (outs),
4431 (ins i32mem:$dst, VR128:$src),
4432 "movd\t{$src, $dst|$dst, $src}",
4433 [(store (i32 (vector_extract (v4i32 VR128:$src),
4434 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4435 VEX, Sched<[WriteLoad]>;
4436 def MOVPDI2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4437 "movd\t{$src, $dst|$dst, $src}",
4438 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4439 (iPTR 0)))], IIC_SSE_MOVD_ToGP>,
4441 def MOVPDI2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4442 "movd\t{$src, $dst|$dst, $src}",
4443 [(store (i32 (vector_extract (v4i32 VR128:$src),
4444 (iPTR 0))), addr:$dst)],
4445 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4447 def : Pat<(v8i32 (X86Vinsert (v8i32 immAllZerosV), GR32:$src2, (iPTR 0))),
4448 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4450 def : Pat<(v4i64 (X86Vinsert (bc_v4i64 (v8i32 immAllZerosV)), GR64:$src2, (iPTR 0))),
4451 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4453 def : Pat<(v8i32 (X86Vinsert undef, GR32:$src2, (iPTR 0))),
4454 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4456 def : Pat<(v4i64 (X86Vinsert undef, GR64:$src2, (iPTR 0))),
4457 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4459 //===---------------------------------------------------------------------===//
4460 // Move Packed Doubleword Int first element to Doubleword Int
4462 let SchedRW = [WriteMove] in {
4463 def VMOVPQIto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4464 "mov{d|q}\t{$src, $dst|$dst, $src}",
4465 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4470 def MOVPQIto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4471 "mov{d|q}\t{$src, $dst|$dst, $src}",
4472 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4477 //===---------------------------------------------------------------------===//
4478 // Bitcast FR64 <-> GR64
4480 let Predicates = [UseAVX] in
4481 def VMOV64toSDrm : VS2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4482 "vmovq\t{$src, $dst|$dst, $src}",
4483 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4484 VEX, Sched<[WriteLoad]>;
4485 def VMOVSDto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4486 "mov{d|q}\t{$src, $dst|$dst, $src}",
4487 [(set GR64:$dst, (bitconvert FR64:$src))],
4488 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4489 def VMOVSDto64mr : VRS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4490 "movq\t{$src, $dst|$dst, $src}",
4491 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4492 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4494 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4495 "movq\t{$src, $dst|$dst, $src}",
4496 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4497 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4498 def MOVSDto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4499 "mov{d|q}\t{$src, $dst|$dst, $src}",
4500 [(set GR64:$dst, (bitconvert FR64:$src))],
4501 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4502 def MOVSDto64mr : RS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4503 "movq\t{$src, $dst|$dst, $src}",
4504 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4505 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4507 //===---------------------------------------------------------------------===//
4508 // Move Scalar Single to Double Int
4510 def VMOVSS2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4511 "movd\t{$src, $dst|$dst, $src}",
4512 [(set GR32:$dst, (bitconvert FR32:$src))],
4513 IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>;
4514 def VMOVSS2DImr : VS2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4515 "movd\t{$src, $dst|$dst, $src}",
4516 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4517 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4518 def MOVSS2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4519 "movd\t{$src, $dst|$dst, $src}",
4520 [(set GR32:$dst, (bitconvert FR32:$src))],
4521 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4522 def MOVSS2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4523 "movd\t{$src, $dst|$dst, $src}",
4524 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4525 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4527 //===---------------------------------------------------------------------===//
4528 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4530 let SchedRW = [WriteMove] in {
4531 let AddedComplexity = 15 in {
4532 def VMOVZDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4533 "movd\t{$src, $dst|$dst, $src}",
4534 [(set VR128:$dst, (v4i32 (X86vzmovl
4535 (v4i32 (scalar_to_vector GR32:$src)))))],
4536 IIC_SSE_MOVDQ>, VEX;
4537 def VMOVZQI2PQIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4538 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4539 [(set VR128:$dst, (v2i64 (X86vzmovl
4540 (v2i64 (scalar_to_vector GR64:$src)))))],
4544 let AddedComplexity = 15 in {
4545 def MOVZDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4546 "movd\t{$src, $dst|$dst, $src}",
4547 [(set VR128:$dst, (v4i32 (X86vzmovl
4548 (v4i32 (scalar_to_vector GR32:$src)))))],
4550 def MOVZQI2PQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4551 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4552 [(set VR128:$dst, (v2i64 (X86vzmovl
4553 (v2i64 (scalar_to_vector GR64:$src)))))],
4558 let AddedComplexity = 20, SchedRW = [WriteLoad] in {
4559 def VMOVZDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4560 "movd\t{$src, $dst|$dst, $src}",
4562 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4563 (loadi32 addr:$src))))))],
4564 IIC_SSE_MOVDQ>, VEX;
4565 def MOVZDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4566 "movd\t{$src, $dst|$dst, $src}",
4568 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4569 (loadi32 addr:$src))))))],
4571 } // AddedComplexity, SchedRW
4573 let Predicates = [UseAVX] in {
4574 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4575 let AddedComplexity = 20 in {
4576 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4577 (VMOVZDI2PDIrm addr:$src)>;
4578 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4579 (VMOVZDI2PDIrm addr:$src)>;
4581 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4582 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4583 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4584 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4585 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4586 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4587 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4590 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4591 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4592 (MOVZDI2PDIrm addr:$src)>;
4593 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4594 (MOVZDI2PDIrm addr:$src)>;
4597 // These are the correct encodings of the instructions so that we know how to
4598 // read correct assembly, even though we continue to emit the wrong ones for
4599 // compatibility with Darwin's buggy assembler.
4600 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4601 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4602 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4603 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4604 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4605 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4606 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4607 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4608 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4609 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4610 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4611 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4613 //===---------------------------------------------------------------------===//
4614 // SSE2 - Move Quadword
4615 //===---------------------------------------------------------------------===//
4617 //===---------------------------------------------------------------------===//
4618 // Move Quadword Int to Packed Quadword Int
4621 let SchedRW = [WriteLoad] in {
4622 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4623 "vmovq\t{$src, $dst|$dst, $src}",
4625 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4626 VEX, Requires<[UseAVX]>;
4627 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4628 "movq\t{$src, $dst|$dst, $src}",
4630 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4632 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4635 //===---------------------------------------------------------------------===//
4636 // Move Packed Quadword Int to Quadword Int
4638 let SchedRW = [WriteStore] in {
4639 def VMOVPQI2QImr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4640 "movq\t{$src, $dst|$dst, $src}",
4641 [(store (i64 (vector_extract (v2i64 VR128:$src),
4642 (iPTR 0))), addr:$dst)],
4643 IIC_SSE_MOVDQ>, VEX;
4644 def MOVPQI2QImr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4645 "movq\t{$src, $dst|$dst, $src}",
4646 [(store (i64 (vector_extract (v2i64 VR128:$src),
4647 (iPTR 0))), addr:$dst)],
4651 //===---------------------------------------------------------------------===//
4652 // Store / copy lower 64-bits of a XMM register.
4654 def VMOVLQ128mr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4655 "movq\t{$src, $dst|$dst, $src}",
4656 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX,
4657 Sched<[WriteStore]>;
4658 def MOVLQ128mr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4659 "movq\t{$src, $dst|$dst, $src}",
4660 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4661 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4663 let AddedComplexity = 20 in
4664 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4665 "vmovq\t{$src, $dst|$dst, $src}",
4667 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4668 (loadi64 addr:$src))))))],
4670 XS, VEX, Requires<[UseAVX]>, Sched<[WriteLoad]>;
4672 let AddedComplexity = 20 in
4673 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4674 "movq\t{$src, $dst|$dst, $src}",
4676 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4677 (loadi64 addr:$src))))))],
4679 XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>;
4681 let Predicates = [UseAVX], AddedComplexity = 20 in {
4682 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4683 (VMOVZQI2PQIrm addr:$src)>;
4684 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4685 (VMOVZQI2PQIrm addr:$src)>;
4686 def : Pat<(v2i64 (X86vzload addr:$src)),
4687 (VMOVZQI2PQIrm addr:$src)>;
4690 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4691 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4692 (MOVZQI2PQIrm addr:$src)>;
4693 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4694 (MOVZQI2PQIrm addr:$src)>;
4695 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4698 let Predicates = [HasAVX] in {
4699 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4700 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4701 def : Pat<(v4i64 (X86vzload addr:$src)),
4702 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4705 //===---------------------------------------------------------------------===//
4706 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4707 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4709 let SchedRW = [WriteVecLogic] in {
4710 let AddedComplexity = 15 in
4711 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4712 "vmovq\t{$src, $dst|$dst, $src}",
4713 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4715 XS, VEX, Requires<[UseAVX]>;
4716 let AddedComplexity = 15 in
4717 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4718 "movq\t{$src, $dst|$dst, $src}",
4719 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4721 XS, Requires<[UseSSE2]>;
4724 let SchedRW = [WriteVecLogicLd] in {
4725 let AddedComplexity = 20 in
4726 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4727 "vmovq\t{$src, $dst|$dst, $src}",
4728 [(set VR128:$dst, (v2i64 (X86vzmovl
4729 (loadv2i64 addr:$src))))],
4731 XS, VEX, Requires<[UseAVX]>;
4732 let AddedComplexity = 20 in {
4733 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4734 "movq\t{$src, $dst|$dst, $src}",
4735 [(set VR128:$dst, (v2i64 (X86vzmovl
4736 (loadv2i64 addr:$src))))],
4738 XS, Requires<[UseSSE2]>;
4742 let AddedComplexity = 20 in {
4743 let Predicates = [UseAVX] in {
4744 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4745 (VMOVZPQILo2PQIrm addr:$src)>;
4746 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4747 (VMOVZPQILo2PQIrr VR128:$src)>;
4749 let Predicates = [UseSSE2] in {
4750 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4751 (MOVZPQILo2PQIrm addr:$src)>;
4752 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4753 (MOVZPQILo2PQIrr VR128:$src)>;
4757 // Instructions to match in the assembler
4758 let SchedRW = [WriteMove] in {
4759 def VMOVQs64rr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4760 "movq\t{$src, $dst|$dst, $src}", [],
4761 IIC_SSE_MOVDQ>, VEX, VEX_W;
4762 def VMOVQd64rr : VS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4763 "movq\t{$src, $dst|$dst, $src}", [],
4764 IIC_SSE_MOVDQ>, VEX, VEX_W;
4765 // Recognize "movd" with GR64 destination, but encode as a "movq"
4766 def VMOVQd64rr_alt : VS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4767 "movq\t{$src, $dst|$dst, $src}", [],
4768 IIC_SSE_MOVDQ>, VEX, VEX_W;
4771 // Instructions for the disassembler
4772 // xr = XMM register
4775 let SchedRW = [WriteMove] in {
4776 let Predicates = [UseAVX] in
4777 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4778 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4779 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4780 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4783 //===---------------------------------------------------------------------===//
4784 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4785 //===---------------------------------------------------------------------===//
4786 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4787 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4788 X86MemOperand x86memop> {
4789 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4790 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4791 [(set RC:$dst, (vt (OpNode RC:$src)))],
4792 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4793 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4794 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4795 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4796 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4799 let Predicates = [HasAVX] in {
4800 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4801 v4f32, VR128, memopv4f32, f128mem>, VEX;
4802 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4803 v4f32, VR128, memopv4f32, f128mem>, VEX;
4804 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4805 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4806 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4807 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4809 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4810 memopv4f32, f128mem>;
4811 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4812 memopv4f32, f128mem>;
4814 let Predicates = [HasAVX] in {
4815 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4816 (VMOVSHDUPrr VR128:$src)>;
4817 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4818 (VMOVSHDUPrm addr:$src)>;
4819 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4820 (VMOVSLDUPrr VR128:$src)>;
4821 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4822 (VMOVSLDUPrm addr:$src)>;
4823 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4824 (VMOVSHDUPYrr VR256:$src)>;
4825 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4826 (VMOVSHDUPYrm addr:$src)>;
4827 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4828 (VMOVSLDUPYrr VR256:$src)>;
4829 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4830 (VMOVSLDUPYrm addr:$src)>;
4833 let Predicates = [UseSSE3] in {
4834 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4835 (MOVSHDUPrr VR128:$src)>;
4836 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4837 (MOVSHDUPrm addr:$src)>;
4838 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4839 (MOVSLDUPrr VR128:$src)>;
4840 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4841 (MOVSLDUPrm addr:$src)>;
4844 //===---------------------------------------------------------------------===//
4845 // SSE3 - Replicate Double FP - MOVDDUP
4846 //===---------------------------------------------------------------------===//
4848 multiclass sse3_replicate_dfp<string OpcodeStr> {
4849 let neverHasSideEffects = 1 in
4850 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4851 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4852 [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4853 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4854 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4857 (scalar_to_vector (loadf64 addr:$src)))))],
4858 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4861 // FIXME: Merge with above classe when there're patterns for the ymm version
4862 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4863 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4864 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4865 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>,
4866 Sched<[WriteShuffle]>;
4867 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4868 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4871 (scalar_to_vector (loadf64 addr:$src)))))]>,
4872 Sched<[WriteShuffleLd]>;
4875 let Predicates = [HasAVX] in {
4876 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4877 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4880 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4882 let Predicates = [HasAVX] in {
4883 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4884 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4885 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4886 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4887 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4888 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4889 def : Pat<(X86Movddup (bc_v2f64
4890 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4891 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4894 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4895 (VMOVDDUPYrm addr:$src)>;
4896 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4897 (VMOVDDUPYrm addr:$src)>;
4898 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4899 (VMOVDDUPYrm addr:$src)>;
4900 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4901 (VMOVDDUPYrr VR256:$src)>;
4904 let Predicates = [UseSSE3] in {
4905 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4906 (MOVDDUPrm addr:$src)>;
4907 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4908 (MOVDDUPrm addr:$src)>;
4909 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4910 (MOVDDUPrm addr:$src)>;
4911 def : Pat<(X86Movddup (bc_v2f64
4912 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4913 (MOVDDUPrm addr:$src)>;
4916 //===---------------------------------------------------------------------===//
4917 // SSE3 - Move Unaligned Integer
4918 //===---------------------------------------------------------------------===//
4920 let SchedRW = [WriteLoad] in {
4921 let Predicates = [HasAVX] in {
4922 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4923 "vlddqu\t{$src, $dst|$dst, $src}",
4924 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4925 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4926 "vlddqu\t{$src, $dst|$dst, $src}",
4927 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
4930 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4931 "lddqu\t{$src, $dst|$dst, $src}",
4932 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4936 //===---------------------------------------------------------------------===//
4937 // SSE3 - Arithmetic
4938 //===---------------------------------------------------------------------===//
4940 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4941 X86MemOperand x86memop, OpndItins itins,
4943 def rr : I<0xD0, MRMSrcReg,
4944 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4946 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4947 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4948 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>,
4949 Sched<[itins.Sched]>;
4950 def rm : I<0xD0, MRMSrcMem,
4951 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4953 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4954 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4955 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>,
4956 Sched<[itins.Sched.Folded, ReadAfterLd]>;
4959 let Predicates = [HasAVX] in {
4960 let ExeDomain = SSEPackedSingle in {
4961 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4962 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4963 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4964 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
4966 let ExeDomain = SSEPackedDouble in {
4967 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4968 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4969 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4970 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
4973 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
4974 let ExeDomain = SSEPackedSingle in
4975 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4976 f128mem, SSE_ALU_F32P>, TB, XD;
4977 let ExeDomain = SSEPackedDouble in
4978 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4979 f128mem, SSE_ALU_F64P>, TB, OpSize;
4982 //===---------------------------------------------------------------------===//
4983 // SSE3 Instructions
4984 //===---------------------------------------------------------------------===//
4987 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4988 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4989 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4991 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4992 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4993 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
4996 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4998 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4999 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5000 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5001 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
5003 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5004 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5005 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5007 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5008 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5009 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
5012 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5014 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5015 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5016 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5017 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
5020 let Predicates = [HasAVX] in {
5021 let ExeDomain = SSEPackedSingle in {
5022 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5023 X86fhadd, 0>, VEX_4V;
5024 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5025 X86fhsub, 0>, VEX_4V;
5026 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5027 X86fhadd, 0>, VEX_4V, VEX_L;
5028 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5029 X86fhsub, 0>, VEX_4V, VEX_L;
5031 let ExeDomain = SSEPackedDouble in {
5032 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5033 X86fhadd, 0>, VEX_4V;
5034 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5035 X86fhsub, 0>, VEX_4V;
5036 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5037 X86fhadd, 0>, VEX_4V, VEX_L;
5038 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5039 X86fhsub, 0>, VEX_4V, VEX_L;
5043 let Constraints = "$src1 = $dst" in {
5044 let ExeDomain = SSEPackedSingle in {
5045 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5046 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5048 let ExeDomain = SSEPackedDouble in {
5049 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5050 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5054 //===---------------------------------------------------------------------===//
5055 // SSSE3 - Packed Absolute Instructions
5056 //===---------------------------------------------------------------------===//
5059 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5060 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5061 Intrinsic IntId128> {
5062 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5064 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5065 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5066 OpSize, Sched<[WriteVecALU]>;
5068 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5070 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5073 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5074 OpSize, Sched<[WriteVecALULd]>;
5077 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5078 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5079 Intrinsic IntId256> {
5080 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5082 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5083 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5084 OpSize, Sched<[WriteVecALU]>;
5086 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5088 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5091 (bitconvert (memopv4i64 addr:$src))))]>, OpSize,
5092 Sched<[WriteVecALULd]>;
5095 // Helper fragments to match sext vXi1 to vXiY.
5096 def v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)),
5098 def v8i1sextv8i16 : PatLeaf<(v8i16 (X86vsrai VR128:$src, (i32 15)))>;
5099 def v4i1sextv4i32 : PatLeaf<(v4i32 (X86vsrai VR128:$src, (i32 31)))>;
5100 def v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)),
5102 def v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256:$src, (i32 15)))>;
5103 def v8i1sextv8i32 : PatLeaf<(v8i32 (X86vsrai VR256:$src, (i32 31)))>;
5105 let Predicates = [HasAVX] in {
5106 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5107 int_x86_ssse3_pabs_b_128>, VEX;
5108 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5109 int_x86_ssse3_pabs_w_128>, VEX;
5110 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5111 int_x86_ssse3_pabs_d_128>, VEX;
5114 (bc_v2i64 (v16i1sextv16i8)),
5115 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5116 (VPABSBrr128 VR128:$src)>;
5118 (bc_v2i64 (v8i1sextv8i16)),
5119 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5120 (VPABSWrr128 VR128:$src)>;
5122 (bc_v2i64 (v4i1sextv4i32)),
5123 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5124 (VPABSDrr128 VR128:$src)>;
5127 let Predicates = [HasAVX2] in {
5128 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5129 int_x86_avx2_pabs_b>, VEX, VEX_L;
5130 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5131 int_x86_avx2_pabs_w>, VEX, VEX_L;
5132 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5133 int_x86_avx2_pabs_d>, VEX, VEX_L;
5136 (bc_v4i64 (v32i1sextv32i8)),
5137 (bc_v4i64 (add (v32i8 VR256:$src), (v32i1sextv32i8)))),
5138 (VPABSBrr256 VR256:$src)>;
5140 (bc_v4i64 (v16i1sextv16i16)),
5141 (bc_v4i64 (add (v16i16 VR256:$src), (v16i1sextv16i16)))),
5142 (VPABSWrr256 VR256:$src)>;
5144 (bc_v4i64 (v8i1sextv8i32)),
5145 (bc_v4i64 (add (v8i32 VR256:$src), (v8i1sextv8i32)))),
5146 (VPABSDrr256 VR256:$src)>;
5149 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5150 int_x86_ssse3_pabs_b_128>;
5151 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5152 int_x86_ssse3_pabs_w_128>;
5153 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5154 int_x86_ssse3_pabs_d_128>;
5156 let Predicates = [HasSSSE3] in {
5158 (bc_v2i64 (v16i1sextv16i8)),
5159 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5160 (PABSBrr128 VR128:$src)>;
5162 (bc_v2i64 (v8i1sextv8i16)),
5163 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5164 (PABSWrr128 VR128:$src)>;
5166 (bc_v2i64 (v4i1sextv4i32)),
5167 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5168 (PABSDrr128 VR128:$src)>;
5171 //===---------------------------------------------------------------------===//
5172 // SSSE3 - Packed Binary Operator Instructions
5173 //===---------------------------------------------------------------------===//
5175 let Sched = WriteVecALU in {
5176 def SSE_PHADDSUBD : OpndItins<
5177 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5179 def SSE_PHADDSUBSW : OpndItins<
5180 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5182 def SSE_PHADDSUBW : OpndItins<
5183 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5186 let Sched = WriteShuffle in
5187 def SSE_PSHUFB : OpndItins<
5188 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5190 let Sched = WriteVecALU in
5191 def SSE_PSIGN : OpndItins<
5192 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5194 let Sched = WriteVecIMul in
5195 def SSE_PMULHRSW : OpndItins<
5196 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5199 /// SS3I_binop_rm - Simple SSSE3 bin op
5200 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5201 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5202 X86MemOperand x86memop, OpndItins itins,
5204 let isCommutable = 1 in
5205 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5206 (ins RC:$src1, RC:$src2),
5208 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5209 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5210 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5211 OpSize, Sched<[itins.Sched]>;
5212 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5213 (ins RC:$src1, x86memop:$src2),
5215 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5216 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5218 (OpVT (OpNode RC:$src1,
5219 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize,
5220 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5223 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5224 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5225 Intrinsic IntId128, OpndItins itins,
5227 let isCommutable = 1 in
5228 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5229 (ins VR128:$src1, VR128:$src2),
5231 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5232 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5233 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5234 OpSize, Sched<[itins.Sched]>;
5235 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5236 (ins VR128:$src1, i128mem:$src2),
5238 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5239 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5241 (IntId128 VR128:$src1,
5242 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize,
5243 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5246 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5247 Intrinsic IntId256> {
5248 let isCommutable = 1 in
5249 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5250 (ins VR256:$src1, VR256:$src2),
5251 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5252 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5254 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5255 (ins VR256:$src1, i256mem:$src2),
5256 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5258 (IntId256 VR256:$src1,
5259 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5262 let ImmT = NoImm, Predicates = [HasAVX] in {
5263 let isCommutable = 0 in {
5264 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5265 memopv2i64, i128mem,
5266 SSE_PHADDSUBW, 0>, VEX_4V;
5267 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5268 memopv2i64, i128mem,
5269 SSE_PHADDSUBD, 0>, VEX_4V;
5270 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5271 memopv2i64, i128mem,
5272 SSE_PHADDSUBW, 0>, VEX_4V;
5273 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5274 memopv2i64, i128mem,
5275 SSE_PHADDSUBD, 0>, VEX_4V;
5276 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5277 memopv2i64, i128mem,
5278 SSE_PSIGN, 0>, VEX_4V;
5279 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5280 memopv2i64, i128mem,
5281 SSE_PSIGN, 0>, VEX_4V;
5282 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5283 memopv2i64, i128mem,
5284 SSE_PSIGN, 0>, VEX_4V;
5285 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5286 memopv2i64, i128mem,
5287 SSE_PSHUFB, 0>, VEX_4V;
5288 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5289 int_x86_ssse3_phadd_sw_128,
5290 SSE_PHADDSUBSW, 0>, VEX_4V;
5291 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5292 int_x86_ssse3_phsub_sw_128,
5293 SSE_PHADDSUBSW, 0>, VEX_4V;
5294 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5295 int_x86_ssse3_pmadd_ub_sw_128,
5296 SSE_PMADD, 0>, VEX_4V;
5298 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5299 int_x86_ssse3_pmul_hr_sw_128,
5300 SSE_PMULHRSW, 0>, VEX_4V;
5303 let ImmT = NoImm, Predicates = [HasAVX2] in {
5304 let isCommutable = 0 in {
5305 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5306 memopv4i64, i256mem,
5307 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5308 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5309 memopv4i64, i256mem,
5310 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5311 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5312 memopv4i64, i256mem,
5313 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5314 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5315 memopv4i64, i256mem,
5316 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5317 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5318 memopv4i64, i256mem,
5319 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5320 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5321 memopv4i64, i256mem,
5322 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5323 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5324 memopv4i64, i256mem,
5325 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5326 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5327 memopv4i64, i256mem,
5328 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5329 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5330 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5331 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5332 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5333 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5334 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5336 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5337 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5340 // None of these have i8 immediate fields.
5341 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5342 let isCommutable = 0 in {
5343 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5344 memopv2i64, i128mem, SSE_PHADDSUBW>;
5345 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5346 memopv2i64, i128mem, SSE_PHADDSUBD>;
5347 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5348 memopv2i64, i128mem, SSE_PHADDSUBW>;
5349 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5350 memopv2i64, i128mem, SSE_PHADDSUBD>;
5351 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5352 memopv2i64, i128mem, SSE_PSIGN>;
5353 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5354 memopv2i64, i128mem, SSE_PSIGN>;
5355 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5356 memopv2i64, i128mem, SSE_PSIGN>;
5357 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5358 memopv2i64, i128mem, SSE_PSHUFB>;
5359 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5360 int_x86_ssse3_phadd_sw_128,
5362 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5363 int_x86_ssse3_phsub_sw_128,
5365 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5366 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5368 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5369 int_x86_ssse3_pmul_hr_sw_128,
5373 //===---------------------------------------------------------------------===//
5374 // SSSE3 - Packed Align Instruction Patterns
5375 //===---------------------------------------------------------------------===//
5377 multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
5378 let neverHasSideEffects = 1 in {
5379 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5380 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5382 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5384 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5385 [], IIC_SSE_PALIGNR>, OpSize, Sched<[WriteShuffle]>;
5387 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5388 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5390 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5392 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5393 [], IIC_SSE_PALIGNR>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5397 multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
5398 let neverHasSideEffects = 1 in {
5399 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5400 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5402 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5403 []>, OpSize, Sched<[WriteShuffle]>;
5405 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5406 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5408 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5409 []>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5413 let Predicates = [HasAVX] in
5414 defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V;
5415 let Predicates = [HasAVX2] in
5416 defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L;
5417 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5418 defm PALIGN : ssse3_palignr<"palignr">;
5420 let Predicates = [HasAVX2] in {
5421 def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5422 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5423 def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5424 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5425 def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5426 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5427 def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5428 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5431 let Predicates = [HasAVX] in {
5432 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5433 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5434 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5435 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5436 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5437 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5438 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5439 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5442 let Predicates = [UseSSSE3] in {
5443 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5444 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5445 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5446 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5447 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5448 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5449 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5450 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5453 //===---------------------------------------------------------------------===//
5454 // SSSE3 - Thread synchronization
5455 //===---------------------------------------------------------------------===//
5457 let SchedRW = [WriteSystem] in {
5458 let usesCustomInserter = 1 in {
5459 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5460 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5461 Requires<[HasSSE3]>;
5464 let Uses = [EAX, ECX, EDX] in
5465 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5466 TB, Requires<[HasSSE3]>;
5467 let Uses = [ECX, EAX] in
5468 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5469 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5470 TB, Requires<[HasSSE3]>;
5473 def : InstAlias<"mwait\t{%eax, %ecx|ecx, eax}", (MWAITrr)>, Requires<[In32BitMode]>;
5474 def : InstAlias<"mwait\t{%rax, %rcx|rcx, rax}", (MWAITrr)>, Requires<[In64BitMode]>;
5476 def : InstAlias<"monitor\t{%eax, %ecx, %edx|edx, ecx, eax}", (MONITORrrr)>,
5477 Requires<[In32BitMode]>;
5478 def : InstAlias<"monitor\t{%rax, %rcx, %rdx|rdx, rcx, rax}", (MONITORrrr)>,
5479 Requires<[In64BitMode]>;
5481 //===----------------------------------------------------------------------===//
5482 // SSE4.1 - Packed Move with Sign/Zero Extend
5483 //===----------------------------------------------------------------------===//
5485 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5486 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5487 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5488 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5490 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5491 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5493 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5497 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5499 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5500 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5501 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5503 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5504 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5505 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5508 let Predicates = [HasAVX] in {
5509 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5511 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5513 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5515 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5517 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5519 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5523 let Predicates = [HasAVX2] in {
5524 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5525 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5526 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5527 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5528 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5529 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5530 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5531 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5532 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5533 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5534 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5535 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5538 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5539 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5540 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5541 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5542 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5543 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5545 let Predicates = [HasAVX] in {
5546 // Common patterns involving scalar load.
5547 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5548 (VPMOVSXBWrm addr:$src)>;
5549 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5550 (VPMOVSXBWrm addr:$src)>;
5551 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5552 (VPMOVSXBWrm addr:$src)>;
5554 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5555 (VPMOVSXWDrm addr:$src)>;
5556 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5557 (VPMOVSXWDrm addr:$src)>;
5558 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5559 (VPMOVSXWDrm addr:$src)>;
5561 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5562 (VPMOVSXDQrm addr:$src)>;
5563 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5564 (VPMOVSXDQrm addr:$src)>;
5565 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5566 (VPMOVSXDQrm addr:$src)>;
5568 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5569 (VPMOVZXBWrm addr:$src)>;
5570 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5571 (VPMOVZXBWrm addr:$src)>;
5572 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5573 (VPMOVZXBWrm addr:$src)>;
5575 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5576 (VPMOVZXWDrm addr:$src)>;
5577 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5578 (VPMOVZXWDrm addr:$src)>;
5579 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5580 (VPMOVZXWDrm addr:$src)>;
5582 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5583 (VPMOVZXDQrm addr:$src)>;
5584 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5585 (VPMOVZXDQrm addr:$src)>;
5586 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5587 (VPMOVZXDQrm addr:$src)>;
5590 let Predicates = [UseSSE41] in {
5591 // Common patterns involving scalar load.
5592 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5593 (PMOVSXBWrm addr:$src)>;
5594 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5595 (PMOVSXBWrm addr:$src)>;
5596 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5597 (PMOVSXBWrm addr:$src)>;
5599 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5600 (PMOVSXWDrm addr:$src)>;
5601 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5602 (PMOVSXWDrm addr:$src)>;
5603 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5604 (PMOVSXWDrm addr:$src)>;
5606 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5607 (PMOVSXDQrm addr:$src)>;
5608 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5609 (PMOVSXDQrm addr:$src)>;
5610 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5611 (PMOVSXDQrm addr:$src)>;
5613 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5614 (PMOVZXBWrm addr:$src)>;
5615 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5616 (PMOVZXBWrm addr:$src)>;
5617 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5618 (PMOVZXBWrm addr:$src)>;
5620 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5621 (PMOVZXWDrm addr:$src)>;
5622 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5623 (PMOVZXWDrm addr:$src)>;
5624 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5625 (PMOVZXWDrm addr:$src)>;
5627 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5628 (PMOVZXDQrm addr:$src)>;
5629 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5630 (PMOVZXDQrm addr:$src)>;
5631 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5632 (PMOVZXDQrm addr:$src)>;
5635 let Predicates = [HasAVX2] in {
5636 let AddedComplexity = 15 in {
5637 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5638 (VPMOVZXDQYrr VR128:$src)>;
5639 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5640 (VPMOVZXWDYrr VR128:$src)>;
5643 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5644 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5647 let Predicates = [HasAVX] in {
5648 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5649 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5652 let Predicates = [UseSSE41] in {
5653 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5654 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5658 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5659 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5660 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5661 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5663 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5664 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5666 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5670 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5672 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5673 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5674 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5676 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5677 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5679 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5683 let Predicates = [HasAVX] in {
5684 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5686 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5688 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5690 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5694 let Predicates = [HasAVX2] in {
5695 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5696 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5697 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5698 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5699 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5700 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5701 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5702 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5705 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5706 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5707 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5708 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5710 let Predicates = [HasAVX] in {
5711 // Common patterns involving scalar load
5712 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5713 (VPMOVSXBDrm addr:$src)>;
5714 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5715 (VPMOVSXWQrm addr:$src)>;
5717 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5718 (VPMOVZXBDrm addr:$src)>;
5719 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5720 (VPMOVZXWQrm addr:$src)>;
5723 let Predicates = [UseSSE41] in {
5724 // Common patterns involving scalar load
5725 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5726 (PMOVSXBDrm addr:$src)>;
5727 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5728 (PMOVSXWQrm addr:$src)>;
5730 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5731 (PMOVZXBDrm addr:$src)>;
5732 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5733 (PMOVZXWQrm addr:$src)>;
5736 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5737 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5738 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5739 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5741 // Expecting a i16 load any extended to i32 value.
5742 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5743 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5744 [(set VR128:$dst, (IntId (bitconvert
5745 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5749 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5751 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5752 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5753 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5755 // Expecting a i16 load any extended to i32 value.
5756 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5757 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5758 [(set VR256:$dst, (IntId (bitconvert
5759 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5763 let Predicates = [HasAVX] in {
5764 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5766 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5769 let Predicates = [HasAVX2] in {
5770 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5771 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5772 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5773 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5775 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5776 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5778 let Predicates = [HasAVX2] in {
5779 def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5780 def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
5781 def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
5783 def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5784 def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
5786 def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5788 def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
5789 (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5790 def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
5791 (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5792 def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
5793 (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5795 def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
5796 (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5797 def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
5798 (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5800 def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
5801 (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5803 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5804 (VPMOVSXWDYrm addr:$src)>;
5805 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5806 (VPMOVSXDQYrm addr:$src)>;
5808 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5809 (scalar_to_vector (loadi64 addr:$src))))))),
5810 (VPMOVSXBDYrm addr:$src)>;
5811 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5812 (scalar_to_vector (loadf64 addr:$src))))))),
5813 (VPMOVSXBDYrm addr:$src)>;
5815 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5816 (scalar_to_vector (loadi64 addr:$src))))))),
5817 (VPMOVSXWQYrm addr:$src)>;
5818 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5819 (scalar_to_vector (loadf64 addr:$src))))))),
5820 (VPMOVSXWQYrm addr:$src)>;
5822 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5823 (scalar_to_vector (loadi32 addr:$src))))))),
5824 (VPMOVSXBQYrm addr:$src)>;
5827 let Predicates = [HasAVX] in {
5828 // Common patterns involving scalar load
5829 def : Pat<(int_x86_sse41_pmovsxbq
5830 (bitconvert (v4i32 (X86vzmovl
5831 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5832 (VPMOVSXBQrm addr:$src)>;
5834 def : Pat<(int_x86_sse41_pmovzxbq
5835 (bitconvert (v4i32 (X86vzmovl
5836 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5837 (VPMOVZXBQrm addr:$src)>;
5840 let Predicates = [UseSSE41] in {
5841 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
5842 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
5843 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
5845 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5846 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
5848 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5850 // Common patterns involving scalar load
5851 def : Pat<(int_x86_sse41_pmovsxbq
5852 (bitconvert (v4i32 (X86vzmovl
5853 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5854 (PMOVSXBQrm addr:$src)>;
5856 def : Pat<(int_x86_sse41_pmovzxbq
5857 (bitconvert (v4i32 (X86vzmovl
5858 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5859 (PMOVZXBQrm addr:$src)>;
5861 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5862 (scalar_to_vector (loadi64 addr:$src))))))),
5863 (PMOVSXWDrm addr:$src)>;
5864 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5865 (scalar_to_vector (loadf64 addr:$src))))))),
5866 (PMOVSXWDrm addr:$src)>;
5867 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5868 (scalar_to_vector (loadi32 addr:$src))))))),
5869 (PMOVSXBDrm addr:$src)>;
5870 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5871 (scalar_to_vector (loadi32 addr:$src))))))),
5872 (PMOVSXWQrm addr:$src)>;
5873 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5874 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5875 (PMOVSXBQrm addr:$src)>;
5876 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5877 (scalar_to_vector (loadi64 addr:$src))))))),
5878 (PMOVSXDQrm addr:$src)>;
5879 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5880 (scalar_to_vector (loadf64 addr:$src))))))),
5881 (PMOVSXDQrm addr:$src)>;
5882 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5883 (scalar_to_vector (loadi64 addr:$src))))))),
5884 (PMOVSXBWrm addr:$src)>;
5885 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5886 (scalar_to_vector (loadf64 addr:$src))))))),
5887 (PMOVSXBWrm addr:$src)>;
5890 let Predicates = [HasAVX2] in {
5891 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5892 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5893 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5895 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5896 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5898 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5900 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
5901 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5902 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
5903 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5904 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
5905 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5907 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
5908 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5909 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
5910 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5912 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
5913 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5916 let Predicates = [HasAVX] in {
5917 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
5918 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
5919 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
5921 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
5922 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
5924 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
5926 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5927 (VPMOVZXBWrm addr:$src)>;
5928 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5929 (VPMOVZXBWrm addr:$src)>;
5930 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5931 (VPMOVZXBDrm addr:$src)>;
5932 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5933 (VPMOVZXBQrm addr:$src)>;
5935 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5936 (VPMOVZXWDrm addr:$src)>;
5937 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5938 (VPMOVZXWDrm addr:$src)>;
5939 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5940 (VPMOVZXWQrm addr:$src)>;
5942 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5943 (VPMOVZXDQrm addr:$src)>;
5944 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5945 (VPMOVZXDQrm addr:$src)>;
5946 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5947 (VPMOVZXDQrm addr:$src)>;
5949 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
5950 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>;
5951 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>;
5953 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5954 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>;
5956 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5958 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5959 (scalar_to_vector (loadi64 addr:$src))))))),
5960 (VPMOVSXWDrm addr:$src)>;
5961 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5962 (scalar_to_vector (loadi64 addr:$src))))))),
5963 (VPMOVSXDQrm addr:$src)>;
5964 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5965 (scalar_to_vector (loadf64 addr:$src))))))),
5966 (VPMOVSXWDrm addr:$src)>;
5967 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5968 (scalar_to_vector (loadf64 addr:$src))))))),
5969 (VPMOVSXDQrm addr:$src)>;
5970 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5971 (scalar_to_vector (loadi64 addr:$src))))))),
5972 (VPMOVSXBWrm addr:$src)>;
5973 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5974 (scalar_to_vector (loadf64 addr:$src))))))),
5975 (VPMOVSXBWrm addr:$src)>;
5977 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5978 (scalar_to_vector (loadi32 addr:$src))))))),
5979 (VPMOVSXBDrm addr:$src)>;
5980 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5981 (scalar_to_vector (loadi32 addr:$src))))))),
5982 (VPMOVSXWQrm addr:$src)>;
5983 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5984 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5985 (VPMOVSXBQrm addr:$src)>;
5988 let Predicates = [UseSSE41] in {
5989 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
5990 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
5991 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
5993 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
5994 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
5996 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
5998 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5999 (PMOVZXBWrm addr:$src)>;
6000 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6001 (PMOVZXBWrm addr:$src)>;
6002 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6003 (PMOVZXBDrm addr:$src)>;
6004 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
6005 (PMOVZXBQrm addr:$src)>;
6007 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6008 (PMOVZXWDrm addr:$src)>;
6009 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6010 (PMOVZXWDrm addr:$src)>;
6011 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6012 (PMOVZXWQrm addr:$src)>;
6014 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6015 (PMOVZXDQrm addr:$src)>;
6016 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6017 (PMOVZXDQrm addr:$src)>;
6018 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
6019 (PMOVZXDQrm addr:$src)>;
6022 //===----------------------------------------------------------------------===//
6023 // SSE4.1 - Extract Instructions
6024 //===----------------------------------------------------------------------===//
6026 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
6027 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
6028 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6029 (ins VR128:$src1, i32i8imm:$src2),
6030 !strconcat(OpcodeStr,
6031 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6032 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
6034 let neverHasSideEffects = 1, mayStore = 1 in
6035 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6036 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
6037 !strconcat(OpcodeStr,
6038 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6041 // There's an AssertZext in the way of writing the store pattern
6042 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6045 let Predicates = [HasAVX] in {
6046 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
6047 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
6048 (ins VR128:$src1, i32i8imm:$src2),
6049 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
6052 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
6055 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
6056 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
6057 let neverHasSideEffects = 1, mayStore = 1 in
6058 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6059 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
6060 !strconcat(OpcodeStr,
6061 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6064 // There's an AssertZext in the way of writing the store pattern
6065 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6068 let Predicates = [HasAVX] in
6069 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
6071 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
6074 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6075 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
6076 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6077 (ins VR128:$src1, i32i8imm:$src2),
6078 !strconcat(OpcodeStr,
6079 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6081 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
6082 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6083 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
6084 !strconcat(OpcodeStr,
6085 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6086 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
6087 addr:$dst)]>, OpSize;
6090 let Predicates = [HasAVX] in
6091 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
6093 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
6095 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6096 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
6097 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
6098 (ins VR128:$src1, i32i8imm:$src2),
6099 !strconcat(OpcodeStr,
6100 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6102 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
6103 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6104 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
6105 !strconcat(OpcodeStr,
6106 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6107 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
6108 addr:$dst)]>, OpSize, REX_W;
6111 let Predicates = [HasAVX] in
6112 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
6114 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
6116 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
6118 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
6119 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6120 (ins VR128:$src1, i32i8imm:$src2),
6121 !strconcat(OpcodeStr,
6122 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6124 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
6126 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6127 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
6128 !strconcat(OpcodeStr,
6129 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6130 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
6131 addr:$dst)]>, OpSize;
6134 let ExeDomain = SSEPackedSingle in {
6135 let Predicates = [UseAVX] in {
6136 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
6137 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
6138 (ins VR128:$src1, i32i8imm:$src2),
6139 "vextractps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6142 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
6145 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
6146 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6149 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6151 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6154 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6155 Requires<[UseSSE41]>;
6157 //===----------------------------------------------------------------------===//
6158 // SSE4.1 - Insert Instructions
6159 //===----------------------------------------------------------------------===//
6161 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
6162 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6163 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6165 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6167 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6169 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
6170 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6171 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
6173 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6175 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6177 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
6178 imm:$src3))]>, OpSize;
6181 let Predicates = [HasAVX] in
6182 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
6183 let Constraints = "$src1 = $dst" in
6184 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6186 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6187 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6188 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6190 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6192 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6194 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6196 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6197 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6199 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6201 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6203 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6204 imm:$src3)))]>, OpSize;
6207 let Predicates = [HasAVX] in
6208 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6209 let Constraints = "$src1 = $dst" in
6210 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6212 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6213 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6214 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6216 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6218 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6220 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6222 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6223 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6225 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6227 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6229 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6230 imm:$src3)))]>, OpSize;
6233 let Predicates = [HasAVX] in
6234 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6235 let Constraints = "$src1 = $dst" in
6236 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6238 // insertps has a few different modes, there's the first two here below which
6239 // are optimized inserts that won't zero arbitrary elements in the destination
6240 // vector. The next one matches the intrinsic and could zero arbitrary elements
6241 // in the target vector.
6242 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6243 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6244 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6246 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6248 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6250 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6252 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6253 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6255 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6257 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6259 (X86insrtps VR128:$src1,
6260 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6261 imm:$src3))]>, OpSize;
6264 let ExeDomain = SSEPackedSingle in {
6265 let Predicates = [HasAVX] in
6266 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6267 let Constraints = "$src1 = $dst" in
6268 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6271 //===----------------------------------------------------------------------===//
6272 // SSE4.1 - Round Instructions
6273 //===----------------------------------------------------------------------===//
6275 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6276 X86MemOperand x86memop, RegisterClass RC,
6277 PatFrag mem_frag32, PatFrag mem_frag64,
6278 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6279 let ExeDomain = SSEPackedSingle in {
6280 // Intrinsic operation, reg.
6281 // Vector intrinsic operation, reg
6282 def PSr : SS4AIi8<opcps, MRMSrcReg,
6283 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6284 !strconcat(OpcodeStr,
6285 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6286 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6289 // Vector intrinsic operation, mem
6290 def PSm : SS4AIi8<opcps, MRMSrcMem,
6291 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6292 !strconcat(OpcodeStr,
6293 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6295 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6297 } // ExeDomain = SSEPackedSingle
6299 let ExeDomain = SSEPackedDouble in {
6300 // Vector intrinsic operation, reg
6301 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6302 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6303 !strconcat(OpcodeStr,
6304 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6305 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6308 // Vector intrinsic operation, mem
6309 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6310 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6311 !strconcat(OpcodeStr,
6312 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6314 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6316 } // ExeDomain = SSEPackedDouble
6319 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6322 Intrinsic F64Int, bit Is2Addr = 1> {
6323 let ExeDomain = GenericDomain in {
6325 let hasSideEffects = 0 in
6326 def SSr : SS4AIi8<opcss, MRMSrcReg,
6327 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6329 !strconcat(OpcodeStr,
6330 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6331 !strconcat(OpcodeStr,
6332 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6335 // Intrinsic operation, reg.
6336 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6337 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6339 !strconcat(OpcodeStr,
6340 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6341 !strconcat(OpcodeStr,
6342 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6343 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6346 // Intrinsic operation, mem.
6347 def SSm : SS4AIi8<opcss, MRMSrcMem,
6348 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6350 !strconcat(OpcodeStr,
6351 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6352 !strconcat(OpcodeStr,
6353 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6355 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6359 let hasSideEffects = 0 in
6360 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6361 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6363 !strconcat(OpcodeStr,
6364 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6365 !strconcat(OpcodeStr,
6366 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6369 // Intrinsic operation, reg.
6370 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6371 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6373 !strconcat(OpcodeStr,
6374 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6375 !strconcat(OpcodeStr,
6376 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6377 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6380 // Intrinsic operation, mem.
6381 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6382 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6384 !strconcat(OpcodeStr,
6385 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6386 !strconcat(OpcodeStr,
6387 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6389 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6391 } // ExeDomain = GenericDomain
6394 // FP round - roundss, roundps, roundsd, roundpd
6395 let Predicates = [HasAVX] in {
6397 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6398 memopv4f32, memopv2f64,
6399 int_x86_sse41_round_ps,
6400 int_x86_sse41_round_pd>, VEX;
6401 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6402 memopv8f32, memopv4f64,
6403 int_x86_avx_round_ps_256,
6404 int_x86_avx_round_pd_256>, VEX, VEX_L;
6405 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6406 int_x86_sse41_round_ss,
6407 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6409 def : Pat<(ffloor FR32:$src),
6410 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6411 def : Pat<(f64 (ffloor FR64:$src)),
6412 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6413 def : Pat<(f32 (fnearbyint FR32:$src)),
6414 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6415 def : Pat<(f64 (fnearbyint FR64:$src)),
6416 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6417 def : Pat<(f32 (fceil FR32:$src)),
6418 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6419 def : Pat<(f64 (fceil FR64:$src)),
6420 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6421 def : Pat<(f32 (frint FR32:$src)),
6422 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6423 def : Pat<(f64 (frint FR64:$src)),
6424 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6425 def : Pat<(f32 (ftrunc FR32:$src)),
6426 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6427 def : Pat<(f64 (ftrunc FR64:$src)),
6428 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6430 def : Pat<(v4f32 (ffloor VR128:$src)),
6431 (VROUNDPSr VR128:$src, (i32 0x1))>;
6432 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6433 (VROUNDPSr VR128:$src, (i32 0xC))>;
6434 def : Pat<(v4f32 (fceil VR128:$src)),
6435 (VROUNDPSr VR128:$src, (i32 0x2))>;
6436 def : Pat<(v4f32 (frint VR128:$src)),
6437 (VROUNDPSr VR128:$src, (i32 0x4))>;
6438 def : Pat<(v4f32 (ftrunc VR128:$src)),
6439 (VROUNDPSr VR128:$src, (i32 0x3))>;
6441 def : Pat<(v2f64 (ffloor VR128:$src)),
6442 (VROUNDPDr VR128:$src, (i32 0x1))>;
6443 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6444 (VROUNDPDr VR128:$src, (i32 0xC))>;
6445 def : Pat<(v2f64 (fceil VR128:$src)),
6446 (VROUNDPDr VR128:$src, (i32 0x2))>;
6447 def : Pat<(v2f64 (frint VR128:$src)),
6448 (VROUNDPDr VR128:$src, (i32 0x4))>;
6449 def : Pat<(v2f64 (ftrunc VR128:$src)),
6450 (VROUNDPDr VR128:$src, (i32 0x3))>;
6452 def : Pat<(v8f32 (ffloor VR256:$src)),
6453 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6454 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6455 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6456 def : Pat<(v8f32 (fceil VR256:$src)),
6457 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6458 def : Pat<(v8f32 (frint VR256:$src)),
6459 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6460 def : Pat<(v8f32 (ftrunc VR256:$src)),
6461 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6463 def : Pat<(v4f64 (ffloor VR256:$src)),
6464 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6465 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6466 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6467 def : Pat<(v4f64 (fceil VR256:$src)),
6468 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6469 def : Pat<(v4f64 (frint VR256:$src)),
6470 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6471 def : Pat<(v4f64 (ftrunc VR256:$src)),
6472 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6475 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6476 memopv4f32, memopv2f64,
6477 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6478 let Constraints = "$src1 = $dst" in
6479 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6480 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6482 let Predicates = [UseSSE41] in {
6483 def : Pat<(ffloor FR32:$src),
6484 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6485 def : Pat<(f64 (ffloor FR64:$src)),
6486 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6487 def : Pat<(f32 (fnearbyint FR32:$src)),
6488 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6489 def : Pat<(f64 (fnearbyint FR64:$src)),
6490 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6491 def : Pat<(f32 (fceil FR32:$src)),
6492 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6493 def : Pat<(f64 (fceil FR64:$src)),
6494 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6495 def : Pat<(f32 (frint FR32:$src)),
6496 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6497 def : Pat<(f64 (frint FR64:$src)),
6498 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6499 def : Pat<(f32 (ftrunc FR32:$src)),
6500 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6501 def : Pat<(f64 (ftrunc FR64:$src)),
6502 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6504 def : Pat<(v4f32 (ffloor VR128:$src)),
6505 (ROUNDPSr VR128:$src, (i32 0x1))>;
6506 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6507 (ROUNDPSr VR128:$src, (i32 0xC))>;
6508 def : Pat<(v4f32 (fceil VR128:$src)),
6509 (ROUNDPSr VR128:$src, (i32 0x2))>;
6510 def : Pat<(v4f32 (frint VR128:$src)),
6511 (ROUNDPSr VR128:$src, (i32 0x4))>;
6512 def : Pat<(v4f32 (ftrunc VR128:$src)),
6513 (ROUNDPSr VR128:$src, (i32 0x3))>;
6515 def : Pat<(v2f64 (ffloor VR128:$src)),
6516 (ROUNDPDr VR128:$src, (i32 0x1))>;
6517 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6518 (ROUNDPDr VR128:$src, (i32 0xC))>;
6519 def : Pat<(v2f64 (fceil VR128:$src)),
6520 (ROUNDPDr VR128:$src, (i32 0x2))>;
6521 def : Pat<(v2f64 (frint VR128:$src)),
6522 (ROUNDPDr VR128:$src, (i32 0x4))>;
6523 def : Pat<(v2f64 (ftrunc VR128:$src)),
6524 (ROUNDPDr VR128:$src, (i32 0x3))>;
6527 //===----------------------------------------------------------------------===//
6528 // SSE4.1 - Packed Bit Test
6529 //===----------------------------------------------------------------------===//
6531 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6532 // the intel intrinsic that corresponds to this.
6533 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6534 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6535 "vptest\t{$src2, $src1|$src1, $src2}",
6536 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6538 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6539 "vptest\t{$src2, $src1|$src1, $src2}",
6540 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6543 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6544 "vptest\t{$src2, $src1|$src1, $src2}",
6545 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6547 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6548 "vptest\t{$src2, $src1|$src1, $src2}",
6549 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6553 let Defs = [EFLAGS] in {
6554 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6555 "ptest\t{$src2, $src1|$src1, $src2}",
6556 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6558 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6559 "ptest\t{$src2, $src1|$src1, $src2}",
6560 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6564 // The bit test instructions below are AVX only
6565 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6566 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6567 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6568 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6569 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6570 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6571 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6572 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6576 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6577 let ExeDomain = SSEPackedSingle in {
6578 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6579 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>,
6582 let ExeDomain = SSEPackedDouble in {
6583 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6584 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>,
6589 //===----------------------------------------------------------------------===//
6590 // SSE4.1 - Misc Instructions
6591 //===----------------------------------------------------------------------===//
6593 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6594 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6595 "popcnt{w}\t{$src, $dst|$dst, $src}",
6596 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6598 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6599 "popcnt{w}\t{$src, $dst|$dst, $src}",
6600 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6601 (implicit EFLAGS)]>, OpSize, XS;
6603 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6604 "popcnt{l}\t{$src, $dst|$dst, $src}",
6605 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6607 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6608 "popcnt{l}\t{$src, $dst|$dst, $src}",
6609 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6610 (implicit EFLAGS)]>, XS;
6612 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6613 "popcnt{q}\t{$src, $dst|$dst, $src}",
6614 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6616 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6617 "popcnt{q}\t{$src, $dst|$dst, $src}",
6618 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6619 (implicit EFLAGS)]>, XS;
6624 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6625 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6626 Intrinsic IntId128> {
6627 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6629 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6630 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6631 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6633 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6636 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6639 let Predicates = [HasAVX] in
6640 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6641 int_x86_sse41_phminposuw>, VEX;
6642 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6643 int_x86_sse41_phminposuw>;
6645 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6646 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6647 Intrinsic IntId128, bit Is2Addr = 1> {
6648 let isCommutable = 1 in
6649 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6650 (ins VR128:$src1, VR128:$src2),
6652 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6653 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6654 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6655 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6656 (ins VR128:$src1, i128mem:$src2),
6658 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6659 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6661 (IntId128 VR128:$src1,
6662 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6665 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6666 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6667 Intrinsic IntId256> {
6668 let isCommutable = 1 in
6669 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6670 (ins VR256:$src1, VR256:$src2),
6671 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6672 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6673 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6674 (ins VR256:$src1, i256mem:$src2),
6675 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6677 (IntId256 VR256:$src1,
6678 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6682 /// SS48I_binop_rm - Simple SSE41 binary operator.
6683 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6684 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6685 X86MemOperand x86memop, bit Is2Addr = 1> {
6686 let isCommutable = 1 in
6687 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6688 (ins RC:$src1, RC:$src2),
6690 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6691 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6692 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6693 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6694 (ins RC:$src1, x86memop:$src2),
6696 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6697 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6699 (OpVT (OpNode RC:$src1,
6700 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6703 let Predicates = [HasAVX] in {
6704 let isCommutable = 0 in
6705 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6707 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6708 memopv2i64, i128mem, 0>, VEX_4V;
6709 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6710 memopv2i64, i128mem, 0>, VEX_4V;
6711 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6712 memopv2i64, i128mem, 0>, VEX_4V;
6713 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6714 memopv2i64, i128mem, 0>, VEX_4V;
6715 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6716 memopv2i64, i128mem, 0>, VEX_4V;
6717 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6718 memopv2i64, i128mem, 0>, VEX_4V;
6719 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6720 memopv2i64, i128mem, 0>, VEX_4V;
6721 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6722 memopv2i64, i128mem, 0>, VEX_4V;
6723 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6727 let Predicates = [HasAVX2] in {
6728 let isCommutable = 0 in
6729 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6730 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6731 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6732 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6733 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6734 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6735 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6736 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6737 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6738 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6739 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6740 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6741 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6742 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6743 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6744 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6745 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6746 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6747 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6748 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6751 let Constraints = "$src1 = $dst" in {
6752 let isCommutable = 0 in
6753 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6754 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6755 memopv2i64, i128mem>;
6756 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6757 memopv2i64, i128mem>;
6758 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6759 memopv2i64, i128mem>;
6760 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6761 memopv2i64, i128mem>;
6762 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6763 memopv2i64, i128mem>;
6764 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6765 memopv2i64, i128mem>;
6766 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6767 memopv2i64, i128mem>;
6768 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6769 memopv2i64, i128mem>;
6770 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6773 let Predicates = [HasAVX] in {
6774 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6775 memopv2i64, i128mem, 0>, VEX_4V;
6776 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6777 memopv2i64, i128mem, 0>, VEX_4V;
6779 let Predicates = [HasAVX2] in {
6780 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6781 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6782 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6783 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6786 let Constraints = "$src1 = $dst" in {
6787 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6788 memopv2i64, i128mem>;
6789 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6790 memopv2i64, i128mem>;
6793 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6794 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6795 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6796 X86MemOperand x86memop, bit Is2Addr = 1> {
6797 let isCommutable = 1 in
6798 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6799 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6801 !strconcat(OpcodeStr,
6802 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6803 !strconcat(OpcodeStr,
6804 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6805 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6807 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6808 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6810 !strconcat(OpcodeStr,
6811 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6812 !strconcat(OpcodeStr,
6813 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6816 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6820 let Predicates = [HasAVX] in {
6821 let isCommutable = 0 in {
6822 let ExeDomain = SSEPackedSingle in {
6823 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6824 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6825 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6826 int_x86_avx_blend_ps_256, VR256, memopv8f32,
6827 f256mem, 0>, VEX_4V, VEX_L;
6829 let ExeDomain = SSEPackedDouble in {
6830 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6831 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6832 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6833 int_x86_avx_blend_pd_256,VR256, memopv4f64,
6834 f256mem, 0>, VEX_4V, VEX_L;
6836 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6837 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6838 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6839 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6841 let ExeDomain = SSEPackedSingle in
6842 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6843 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6844 let ExeDomain = SSEPackedDouble in
6845 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6846 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6847 let ExeDomain = SSEPackedSingle in
6848 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6849 VR256, memopv8f32, i256mem, 0>, VEX_4V, VEX_L;
6852 let Predicates = [HasAVX2] in {
6853 let isCommutable = 0 in {
6854 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6855 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6856 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6857 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6861 let Constraints = "$src1 = $dst" in {
6862 let isCommutable = 0 in {
6863 let ExeDomain = SSEPackedSingle in
6864 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6865 VR128, memopv4f32, f128mem>;
6866 let ExeDomain = SSEPackedDouble in
6867 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6868 VR128, memopv2f64, f128mem>;
6869 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6870 VR128, memopv2i64, i128mem>;
6871 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6872 VR128, memopv2i64, i128mem>;
6874 let ExeDomain = SSEPackedSingle in
6875 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6876 VR128, memopv4f32, f128mem>;
6877 let ExeDomain = SSEPackedDouble in
6878 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6879 VR128, memopv2f64, f128mem>;
6882 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6883 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6884 RegisterClass RC, X86MemOperand x86memop,
6885 PatFrag mem_frag, Intrinsic IntId> {
6886 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6887 (ins RC:$src1, RC:$src2, RC:$src3),
6888 !strconcat(OpcodeStr,
6889 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6890 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6891 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6893 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6894 (ins RC:$src1, x86memop:$src2, RC:$src3),
6895 !strconcat(OpcodeStr,
6896 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6898 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6900 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6903 let Predicates = [HasAVX] in {
6904 let ExeDomain = SSEPackedDouble in {
6905 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6906 memopv2f64, int_x86_sse41_blendvpd>;
6907 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6908 memopv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
6909 } // ExeDomain = SSEPackedDouble
6910 let ExeDomain = SSEPackedSingle in {
6911 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6912 memopv4f32, int_x86_sse41_blendvps>;
6913 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6914 memopv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
6915 } // ExeDomain = SSEPackedSingle
6916 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6917 memopv2i64, int_x86_sse41_pblendvb>;
6920 let Predicates = [HasAVX2] in {
6921 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6922 memopv4i64, int_x86_avx2_pblendvb>, VEX_L;
6925 let Predicates = [HasAVX] in {
6926 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6927 (v16i8 VR128:$src2))),
6928 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6929 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6930 (v4i32 VR128:$src2))),
6931 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6932 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6933 (v4f32 VR128:$src2))),
6934 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6935 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6936 (v2i64 VR128:$src2))),
6937 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6938 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6939 (v2f64 VR128:$src2))),
6940 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6941 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6942 (v8i32 VR256:$src2))),
6943 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6944 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6945 (v8f32 VR256:$src2))),
6946 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6947 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6948 (v4i64 VR256:$src2))),
6949 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6950 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6951 (v4f64 VR256:$src2))),
6952 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6954 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6956 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6957 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6959 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6961 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6963 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6964 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6966 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6967 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6969 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6972 let Predicates = [HasAVX2] in {
6973 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6974 (v32i8 VR256:$src2))),
6975 (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
6976 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6978 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6981 /// SS41I_ternary_int - SSE 4.1 ternary operator
6982 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6983 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6984 X86MemOperand x86memop, Intrinsic IntId> {
6985 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6986 (ins VR128:$src1, VR128:$src2),
6987 !strconcat(OpcodeStr,
6988 "\t{$src2, $dst|$dst, $src2}"),
6989 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6992 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6993 (ins VR128:$src1, x86memop:$src2),
6994 !strconcat(OpcodeStr,
6995 "\t{$src2, $dst|$dst, $src2}"),
6998 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
7002 let ExeDomain = SSEPackedDouble in
7003 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
7004 int_x86_sse41_blendvpd>;
7005 let ExeDomain = SSEPackedSingle in
7006 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
7007 int_x86_sse41_blendvps>;
7008 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
7009 int_x86_sse41_pblendvb>;
7011 // Aliases with the implicit xmm0 argument
7012 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7013 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
7014 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7015 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
7016 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7017 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
7018 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7019 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
7020 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7021 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
7022 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7023 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
7025 let Predicates = [UseSSE41] in {
7026 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
7027 (v16i8 VR128:$src2))),
7028 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
7029 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
7030 (v4i32 VR128:$src2))),
7031 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7032 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
7033 (v4f32 VR128:$src2))),
7034 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7035 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
7036 (v2i64 VR128:$src2))),
7037 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7038 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
7039 (v2f64 VR128:$src2))),
7040 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7042 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
7044 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
7045 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
7047 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
7048 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
7050 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7054 let Predicates = [HasAVX] in
7055 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7056 "vmovntdqa\t{$src, $dst|$dst, $src}",
7057 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7059 let Predicates = [HasAVX2] in
7060 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
7061 "vmovntdqa\t{$src, $dst|$dst, $src}",
7062 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
7064 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7065 "movntdqa\t{$src, $dst|$dst, $src}",
7066 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7069 //===----------------------------------------------------------------------===//
7070 // SSE4.2 - Compare Instructions
7071 //===----------------------------------------------------------------------===//
7073 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
7074 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
7075 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
7076 X86MemOperand x86memop, bit Is2Addr = 1> {
7077 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
7078 (ins RC:$src1, RC:$src2),
7080 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7081 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7082 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
7084 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
7085 (ins RC:$src1, x86memop:$src2),
7087 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7088 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7090 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
7093 let Predicates = [HasAVX] in
7094 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
7095 memopv2i64, i128mem, 0>, VEX_4V;
7097 let Predicates = [HasAVX2] in
7098 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
7099 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
7101 let Constraints = "$src1 = $dst" in
7102 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
7103 memopv2i64, i128mem>;
7105 //===----------------------------------------------------------------------===//
7106 // SSE4.2 - String/text Processing Instructions
7107 //===----------------------------------------------------------------------===//
7109 // Packed Compare Implicit Length Strings, Return Mask
7110 multiclass pseudo_pcmpistrm<string asm> {
7111 def REG : PseudoI<(outs VR128:$dst),
7112 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7113 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
7115 def MEM : PseudoI<(outs VR128:$dst),
7116 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7117 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
7118 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7121 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7122 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
7123 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
7126 multiclass pcmpistrm_SS42AI<string asm> {
7127 def rr : SS42AI<0x62, MRMSrcReg, (outs),
7128 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7129 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7132 def rm :SS42AI<0x62, MRMSrcMem, (outs),
7133 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7134 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7138 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
7139 let Predicates = [HasAVX] in
7140 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
7141 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
7144 // Packed Compare Explicit Length Strings, Return Mask
7145 multiclass pseudo_pcmpestrm<string asm> {
7146 def REG : PseudoI<(outs VR128:$dst),
7147 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7148 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
7149 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7150 def MEM : PseudoI<(outs VR128:$dst),
7151 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7152 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
7153 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
7156 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7157 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
7158 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
7161 multiclass SS42AI_pcmpestrm<string asm> {
7162 def rr : SS42AI<0x60, MRMSrcReg, (outs),
7163 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7164 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7167 def rm : SS42AI<0x60, MRMSrcMem, (outs),
7168 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7169 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7173 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7174 let Predicates = [HasAVX] in
7175 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
7176 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
7179 // Packed Compare Implicit Length Strings, Return Index
7180 multiclass pseudo_pcmpistri<string asm> {
7181 def REG : PseudoI<(outs GR32:$dst),
7182 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7183 [(set GR32:$dst, EFLAGS,
7184 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
7185 def MEM : PseudoI<(outs GR32:$dst),
7186 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7187 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
7188 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7191 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7192 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
7193 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
7196 multiclass SS42AI_pcmpistri<string asm> {
7197 def rr : SS42AI<0x63, MRMSrcReg, (outs),
7198 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7199 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7202 def rm : SS42AI<0x63, MRMSrcMem, (outs),
7203 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7204 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7208 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
7209 let Predicates = [HasAVX] in
7210 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
7211 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
7214 // Packed Compare Explicit Length Strings, Return Index
7215 multiclass pseudo_pcmpestri<string asm> {
7216 def REG : PseudoI<(outs GR32:$dst),
7217 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7218 [(set GR32:$dst, EFLAGS,
7219 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7220 def MEM : PseudoI<(outs GR32:$dst),
7221 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7222 [(set GR32:$dst, EFLAGS,
7223 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
7227 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7228 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
7229 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
7232 multiclass SS42AI_pcmpestri<string asm> {
7233 def rr : SS42AI<0x61, MRMSrcReg, (outs),
7234 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7235 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7238 def rm : SS42AI<0x61, MRMSrcMem, (outs),
7239 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7240 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7244 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7245 let Predicates = [HasAVX] in
7246 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
7247 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
7250 //===----------------------------------------------------------------------===//
7251 // SSE4.2 - CRC Instructions
7252 //===----------------------------------------------------------------------===//
7254 // No CRC instructions have AVX equivalents
7256 // crc intrinsic instruction
7257 // This set of instructions are only rm, the only difference is the size
7259 let Constraints = "$src1 = $dst" in {
7260 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
7261 (ins GR32:$src1, i8mem:$src2),
7262 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7264 (int_x86_sse42_crc32_32_8 GR32:$src1,
7265 (load addr:$src2)))]>;
7266 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
7267 (ins GR32:$src1, GR8:$src2),
7268 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7270 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
7271 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7272 (ins GR32:$src1, i16mem:$src2),
7273 "crc32{w}\t{$src2, $src1|$src1, $src2}",
7275 (int_x86_sse42_crc32_32_16 GR32:$src1,
7276 (load addr:$src2)))]>,
7278 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7279 (ins GR32:$src1, GR16:$src2),
7280 "crc32{w}\t{$src2, $src1|$src1, $src2}",
7282 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7284 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7285 (ins GR32:$src1, i32mem:$src2),
7286 "crc32{l}\t{$src2, $src1|$src1, $src2}",
7288 (int_x86_sse42_crc32_32_32 GR32:$src1,
7289 (load addr:$src2)))]>;
7290 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7291 (ins GR32:$src1, GR32:$src2),
7292 "crc32{l}\t{$src2, $src1|$src1, $src2}",
7294 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7295 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7296 (ins GR64:$src1, i8mem:$src2),
7297 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7299 (int_x86_sse42_crc32_64_8 GR64:$src1,
7300 (load addr:$src2)))]>,
7302 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7303 (ins GR64:$src1, GR8:$src2),
7304 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7306 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7308 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7309 (ins GR64:$src1, i64mem:$src2),
7310 "crc32{q}\t{$src2, $src1|$src1, $src2}",
7312 (int_x86_sse42_crc32_64_64 GR64:$src1,
7313 (load addr:$src2)))]>,
7315 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7316 (ins GR64:$src1, GR64:$src2),
7317 "crc32{q}\t{$src2, $src1|$src1, $src2}",
7319 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7323 //===----------------------------------------------------------------------===//
7324 // AES-NI Instructions
7325 //===----------------------------------------------------------------------===//
7327 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7328 Intrinsic IntId128, bit Is2Addr = 1> {
7329 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7330 (ins VR128:$src1, VR128:$src2),
7332 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7333 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7334 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7336 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7337 (ins VR128:$src1, i128mem:$src2),
7339 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7340 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7342 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7345 // Perform One Round of an AES Encryption/Decryption Flow
7346 let Predicates = [HasAVX, HasAES] in {
7347 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7348 int_x86_aesni_aesenc, 0>, VEX_4V;
7349 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7350 int_x86_aesni_aesenclast, 0>, VEX_4V;
7351 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7352 int_x86_aesni_aesdec, 0>, VEX_4V;
7353 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7354 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7357 let Constraints = "$src1 = $dst" in {
7358 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7359 int_x86_aesni_aesenc>;
7360 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7361 int_x86_aesni_aesenclast>;
7362 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7363 int_x86_aesni_aesdec>;
7364 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7365 int_x86_aesni_aesdeclast>;
7368 // Perform the AES InvMixColumn Transformation
7369 let Predicates = [HasAVX, HasAES] in {
7370 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7372 "vaesimc\t{$src1, $dst|$dst, $src1}",
7374 (int_x86_aesni_aesimc VR128:$src1))]>,
7376 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7377 (ins i128mem:$src1),
7378 "vaesimc\t{$src1, $dst|$dst, $src1}",
7379 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7382 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7384 "aesimc\t{$src1, $dst|$dst, $src1}",
7386 (int_x86_aesni_aesimc VR128:$src1))]>,
7388 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7389 (ins i128mem:$src1),
7390 "aesimc\t{$src1, $dst|$dst, $src1}",
7391 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7394 // AES Round Key Generation Assist
7395 let Predicates = [HasAVX, HasAES] in {
7396 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7397 (ins VR128:$src1, i8imm:$src2),
7398 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7400 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7402 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7403 (ins i128mem:$src1, i8imm:$src2),
7404 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7406 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7409 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7410 (ins VR128:$src1, i8imm:$src2),
7411 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7413 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7415 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7416 (ins i128mem:$src1, i8imm:$src2),
7417 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7419 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7422 //===----------------------------------------------------------------------===//
7423 // PCLMUL Instructions
7424 //===----------------------------------------------------------------------===//
7426 // AVX carry-less Multiplication instructions
7427 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7428 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7429 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7431 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7433 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7434 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7435 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7436 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7437 (memopv2i64 addr:$src2), imm:$src3))]>;
7439 // Carry-less Multiplication instructions
7440 let Constraints = "$src1 = $dst" in {
7441 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7442 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7443 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7445 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7447 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7448 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7449 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7450 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7451 (memopv2i64 addr:$src2), imm:$src3))]>;
7452 } // Constraints = "$src1 = $dst"
7455 multiclass pclmul_alias<string asm, int immop> {
7456 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7457 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7459 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7460 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7462 def : InstAlias<!strconcat("vpclmul", asm,
7463 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7464 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7466 def : InstAlias<!strconcat("vpclmul", asm,
7467 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7468 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7470 defm : pclmul_alias<"hqhq", 0x11>;
7471 defm : pclmul_alias<"hqlq", 0x01>;
7472 defm : pclmul_alias<"lqhq", 0x10>;
7473 defm : pclmul_alias<"lqlq", 0x00>;
7475 //===----------------------------------------------------------------------===//
7476 // SSE4A Instructions
7477 //===----------------------------------------------------------------------===//
7479 let Predicates = [HasSSE4A] in {
7481 let Constraints = "$src = $dst" in {
7482 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7483 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7484 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7485 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7486 imm:$idx))]>, TB, OpSize;
7487 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7488 (ins VR128:$src, VR128:$mask),
7489 "extrq\t{$mask, $src|$src, $mask}",
7490 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7491 VR128:$mask))]>, TB, OpSize;
7493 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7494 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7495 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7496 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7497 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7498 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7499 (ins VR128:$src, VR128:$mask),
7500 "insertq\t{$mask, $src|$src, $mask}",
7501 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7502 VR128:$mask))]>, XD;
7505 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7506 "movntss\t{$src, $dst|$dst, $src}",
7507 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7509 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7510 "movntsd\t{$src, $dst|$dst, $src}",
7511 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7514 //===----------------------------------------------------------------------===//
7516 //===----------------------------------------------------------------------===//
7518 //===----------------------------------------------------------------------===//
7519 // VBROADCAST - Load from memory and broadcast to all elements of the
7520 // destination operand
7522 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7523 X86MemOperand x86memop, Intrinsic Int> :
7524 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7525 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7526 [(set RC:$dst, (Int addr:$src))]>, VEX;
7528 // AVX2 adds register forms
7529 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7531 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7532 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7533 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7535 let ExeDomain = SSEPackedSingle in {
7536 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7537 int_x86_avx_vbroadcast_ss>;
7538 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7539 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7541 let ExeDomain = SSEPackedDouble in
7542 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7543 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7544 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7545 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7547 let ExeDomain = SSEPackedSingle in {
7548 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7549 int_x86_avx2_vbroadcast_ss_ps>;
7550 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7551 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7553 let ExeDomain = SSEPackedDouble in
7554 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7555 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7557 let Predicates = [HasAVX2] in
7558 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7559 int_x86_avx2_vbroadcasti128>, VEX_L;
7561 let Predicates = [HasAVX] in
7562 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7563 (VBROADCASTF128 addr:$src)>;
7566 //===----------------------------------------------------------------------===//
7567 // VINSERTF128 - Insert packed floating-point values
7569 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7570 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7571 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7572 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7575 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7576 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7577 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7581 let Predicates = [HasAVX] in {
7582 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7584 (VINSERTF128rr VR256:$src1, VR128:$src2,
7585 (INSERT_get_vinsert128_imm VR256:$ins))>;
7586 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7588 (VINSERTF128rr VR256:$src1, VR128:$src2,
7589 (INSERT_get_vinsert128_imm VR256:$ins))>;
7591 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (memopv4f32 addr:$src2),
7593 (VINSERTF128rm VR256:$src1, addr:$src2,
7594 (INSERT_get_vinsert128_imm VR256:$ins))>;
7595 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (memopv2f64 addr:$src2),
7597 (VINSERTF128rm VR256:$src1, addr:$src2,
7598 (INSERT_get_vinsert128_imm VR256:$ins))>;
7601 let Predicates = [HasAVX1Only] in {
7602 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7604 (VINSERTF128rr VR256:$src1, VR128:$src2,
7605 (INSERT_get_vinsert128_imm VR256:$ins))>;
7606 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7608 (VINSERTF128rr VR256:$src1, VR128:$src2,
7609 (INSERT_get_vinsert128_imm VR256:$ins))>;
7610 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7612 (VINSERTF128rr VR256:$src1, VR128:$src2,
7613 (INSERT_get_vinsert128_imm VR256:$ins))>;
7614 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7616 (VINSERTF128rr VR256:$src1, VR128:$src2,
7617 (INSERT_get_vinsert128_imm VR256:$ins))>;
7619 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
7621 (VINSERTF128rm VR256:$src1, addr:$src2,
7622 (INSERT_get_vinsert128_imm VR256:$ins))>;
7623 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
7624 (bc_v4i32 (memopv2i64 addr:$src2)),
7626 (VINSERTF128rm VR256:$src1, addr:$src2,
7627 (INSERT_get_vinsert128_imm VR256:$ins))>;
7628 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
7629 (bc_v16i8 (memopv2i64 addr:$src2)),
7631 (VINSERTF128rm VR256:$src1, addr:$src2,
7632 (INSERT_get_vinsert128_imm VR256:$ins))>;
7633 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
7634 (bc_v8i16 (memopv2i64 addr:$src2)),
7636 (VINSERTF128rm VR256:$src1, addr:$src2,
7637 (INSERT_get_vinsert128_imm VR256:$ins))>;
7640 //===----------------------------------------------------------------------===//
7641 // VEXTRACTF128 - Extract packed floating-point values
7643 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7644 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7645 (ins VR256:$src1, i8imm:$src2),
7646 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7649 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7650 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7651 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7656 let Predicates = [HasAVX] in {
7657 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7658 (v4f32 (VEXTRACTF128rr
7659 (v8f32 VR256:$src1),
7660 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7661 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7662 (v2f64 (VEXTRACTF128rr
7663 (v4f64 VR256:$src1),
7664 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7666 def : Pat<(alignedstore (v4f32 (vextract128_extract:$ext (v8f32 VR256:$src1),
7667 (iPTR imm))), addr:$dst),
7668 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7669 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7670 def : Pat<(alignedstore (v2f64 (vextract128_extract:$ext (v4f64 VR256:$src1),
7671 (iPTR imm))), addr:$dst),
7672 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7673 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7676 let Predicates = [HasAVX1Only] in {
7677 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7678 (v2i64 (VEXTRACTF128rr
7679 (v4i64 VR256:$src1),
7680 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7681 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7682 (v4i32 (VEXTRACTF128rr
7683 (v8i32 VR256:$src1),
7684 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7685 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7686 (v8i16 (VEXTRACTF128rr
7687 (v16i16 VR256:$src1),
7688 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7689 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7690 (v16i8 (VEXTRACTF128rr
7691 (v32i8 VR256:$src1),
7692 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7694 def : Pat<(alignedstore (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
7695 (iPTR imm))), addr:$dst),
7696 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7697 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7698 def : Pat<(alignedstore (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
7699 (iPTR imm))), addr:$dst),
7700 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7701 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7702 def : Pat<(alignedstore (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
7703 (iPTR imm))), addr:$dst),
7704 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7705 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7706 def : Pat<(alignedstore (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
7707 (iPTR imm))), addr:$dst),
7708 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7709 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7712 //===----------------------------------------------------------------------===//
7713 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7715 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7716 Intrinsic IntLd, Intrinsic IntLd256,
7717 Intrinsic IntSt, Intrinsic IntSt256> {
7718 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7719 (ins VR128:$src1, f128mem:$src2),
7720 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7721 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7723 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7724 (ins VR256:$src1, f256mem:$src2),
7725 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7726 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7728 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7729 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7730 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7731 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7732 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7733 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7734 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7735 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7738 let ExeDomain = SSEPackedSingle in
7739 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7740 int_x86_avx_maskload_ps,
7741 int_x86_avx_maskload_ps_256,
7742 int_x86_avx_maskstore_ps,
7743 int_x86_avx_maskstore_ps_256>;
7744 let ExeDomain = SSEPackedDouble in
7745 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7746 int_x86_avx_maskload_pd,
7747 int_x86_avx_maskload_pd_256,
7748 int_x86_avx_maskstore_pd,
7749 int_x86_avx_maskstore_pd_256>;
7751 //===----------------------------------------------------------------------===//
7752 // VPERMIL - Permute Single and Double Floating-Point Values
7754 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7755 RegisterClass RC, X86MemOperand x86memop_f,
7756 X86MemOperand x86memop_i, PatFrag i_frag,
7757 Intrinsic IntVar, ValueType vt> {
7758 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7759 (ins RC:$src1, RC:$src2),
7760 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7761 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7762 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7763 (ins RC:$src1, x86memop_i:$src2),
7764 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7765 [(set RC:$dst, (IntVar RC:$src1,
7766 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7768 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7769 (ins RC:$src1, i8imm:$src2),
7770 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7771 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7772 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7773 (ins x86memop_f:$src1, i8imm:$src2),
7774 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7776 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7779 let ExeDomain = SSEPackedSingle in {
7780 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7781 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7782 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7783 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7785 let ExeDomain = SSEPackedDouble in {
7786 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7787 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7788 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7789 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7792 let Predicates = [HasAVX] in {
7793 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7794 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7795 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7796 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7797 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7799 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7800 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7801 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7803 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7804 (VPERMILPDri VR128:$src1, imm:$imm)>;
7805 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7806 (VPERMILPDmi addr:$src1, imm:$imm)>;
7809 //===----------------------------------------------------------------------===//
7810 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7812 let ExeDomain = SSEPackedSingle in {
7813 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7814 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7815 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7816 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7817 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7818 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7819 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7820 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7821 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7822 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7825 let Predicates = [HasAVX] in {
7826 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7827 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7828 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7829 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7830 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7833 let Predicates = [HasAVX1Only] in {
7834 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7835 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7836 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7837 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7838 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7839 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7840 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7841 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7843 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7844 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7845 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7846 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7847 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7848 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7849 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7850 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7851 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7852 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7853 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7854 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7857 //===----------------------------------------------------------------------===//
7858 // VZERO - Zero YMM registers
7860 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7861 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7862 // Zero All YMM registers
7863 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7864 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7866 // Zero Upper bits of YMM registers
7867 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7868 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7871 //===----------------------------------------------------------------------===//
7872 // Half precision conversion instructions
7873 //===----------------------------------------------------------------------===//
7874 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7875 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7876 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7877 [(set RC:$dst, (Int VR128:$src))]>,
7879 let neverHasSideEffects = 1, mayLoad = 1 in
7880 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7881 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7884 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7885 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7886 (ins RC:$src1, i32i8imm:$src2),
7887 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7888 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7890 let neverHasSideEffects = 1, mayStore = 1 in
7891 def mr : Ii8<0x1D, MRMDestMem, (outs),
7892 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7893 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7897 let Predicates = [HasAVX, HasF16C] in {
7898 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7899 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
7900 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7901 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
7904 //===----------------------------------------------------------------------===//
7905 // AVX2 Instructions
7906 //===----------------------------------------------------------------------===//
7908 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7909 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7910 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7911 X86MemOperand x86memop> {
7912 let isCommutable = 1 in
7913 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7914 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7915 !strconcat(OpcodeStr,
7916 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7917 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7919 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7920 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7921 !strconcat(OpcodeStr,
7922 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7925 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7929 let isCommutable = 0 in {
7930 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7931 VR128, memopv2i64, i128mem>;
7932 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7933 VR256, memopv4i64, i256mem>, VEX_L;
7936 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
7938 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7939 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
7941 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7943 //===----------------------------------------------------------------------===//
7944 // VPBROADCAST - Load from memory and broadcast to all elements of the
7945 // destination operand
7947 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7948 X86MemOperand x86memop, PatFrag ld_frag,
7949 Intrinsic Int128, Intrinsic Int256> {
7950 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7951 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7952 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7953 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7954 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7956 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7957 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7958 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7959 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
7960 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7961 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7963 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
7967 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7968 int_x86_avx2_pbroadcastb_128,
7969 int_x86_avx2_pbroadcastb_256>;
7970 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7971 int_x86_avx2_pbroadcastw_128,
7972 int_x86_avx2_pbroadcastw_256>;
7973 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7974 int_x86_avx2_pbroadcastd_128,
7975 int_x86_avx2_pbroadcastd_256>;
7976 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7977 int_x86_avx2_pbroadcastq_128,
7978 int_x86_avx2_pbroadcastq_256>;
7980 let Predicates = [HasAVX2] in {
7981 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7982 (VPBROADCASTBrm addr:$src)>;
7983 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7984 (VPBROADCASTBYrm addr:$src)>;
7985 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7986 (VPBROADCASTWrm addr:$src)>;
7987 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7988 (VPBROADCASTWYrm addr:$src)>;
7989 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7990 (VPBROADCASTDrm addr:$src)>;
7991 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7992 (VPBROADCASTDYrm addr:$src)>;
7993 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7994 (VPBROADCASTQrm addr:$src)>;
7995 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7996 (VPBROADCASTQYrm addr:$src)>;
7998 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7999 (VPBROADCASTBrr VR128:$src)>;
8000 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
8001 (VPBROADCASTBYrr VR128:$src)>;
8002 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
8003 (VPBROADCASTWrr VR128:$src)>;
8004 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
8005 (VPBROADCASTWYrr VR128:$src)>;
8006 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
8007 (VPBROADCASTDrr VR128:$src)>;
8008 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
8009 (VPBROADCASTDYrr VR128:$src)>;
8010 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
8011 (VPBROADCASTQrr VR128:$src)>;
8012 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
8013 (VPBROADCASTQYrr VR128:$src)>;
8014 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
8015 (VBROADCASTSSrr VR128:$src)>;
8016 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
8017 (VBROADCASTSSYrr VR128:$src)>;
8018 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
8019 (VPBROADCASTQrr VR128:$src)>;
8020 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
8021 (VBROADCASTSDYrr VR128:$src)>;
8023 // Provide fallback in case the load node that is used in the patterns above
8024 // is used by additional users, which prevents the pattern selection.
8025 let AddedComplexity = 20 in {
8026 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8027 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8028 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8029 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8030 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8031 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
8033 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8034 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8035 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8036 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8037 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8038 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
8042 // AVX1 broadcast patterns
8043 let Predicates = [HasAVX1Only] in {
8044 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8045 (VBROADCASTSSYrm addr:$src)>;
8046 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8047 (VBROADCASTSDYrm addr:$src)>;
8048 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8049 (VBROADCASTSSrm addr:$src)>;
8052 let Predicates = [HasAVX] in {
8053 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
8054 (VBROADCASTSSYrm addr:$src)>;
8055 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
8056 (VBROADCASTSDYrm addr:$src)>;
8057 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
8058 (VBROADCASTSSrm addr:$src)>;
8060 // Provide fallback in case the load node that is used in the patterns above
8061 // is used by additional users, which prevents the pattern selection.
8062 let AddedComplexity = 20 in {
8063 // 128bit broadcasts:
8064 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8065 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
8066 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8067 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
8068 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
8069 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
8070 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8071 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
8072 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
8073 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
8075 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8076 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
8077 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8078 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
8079 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
8080 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
8081 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8082 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
8083 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
8084 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
8088 //===----------------------------------------------------------------------===//
8089 // VPERM - Permute instructions
8092 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8094 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8095 (ins VR256:$src1, VR256:$src2),
8096 !strconcat(OpcodeStr,
8097 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8099 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
8101 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8102 (ins VR256:$src1, i256mem:$src2),
8103 !strconcat(OpcodeStr,
8104 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8106 (OpVT (X86VPermv VR256:$src1,
8107 (bitconvert (mem_frag addr:$src2)))))]>,
8111 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
8112 let ExeDomain = SSEPackedSingle in
8113 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
8115 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8117 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
8118 (ins VR256:$src1, i8imm:$src2),
8119 !strconcat(OpcodeStr,
8120 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8122 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
8124 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
8125 (ins i256mem:$src1, i8imm:$src2),
8126 !strconcat(OpcodeStr,
8127 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8129 (OpVT (X86VPermi (mem_frag addr:$src1),
8130 (i8 imm:$src2))))]>, VEX, VEX_L;
8133 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
8134 let ExeDomain = SSEPackedDouble in
8135 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
8137 //===----------------------------------------------------------------------===//
8138 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
8140 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
8141 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
8142 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8143 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
8144 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
8145 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
8146 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
8147 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8148 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
8149 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
8151 let Predicates = [HasAVX2] in {
8152 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8153 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8154 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8155 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8156 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8157 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8159 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
8161 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8162 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
8163 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
8164 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8165 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
8167 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8171 //===----------------------------------------------------------------------===//
8172 // VINSERTI128 - Insert packed integer values
8174 let neverHasSideEffects = 1 in {
8175 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
8176 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
8177 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8180 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
8181 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
8182 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8186 let Predicates = [HasAVX2] in {
8187 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
8189 (VINSERTI128rr VR256:$src1, VR128:$src2,
8190 (INSERT_get_vinsert128_imm VR256:$ins))>;
8191 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
8193 (VINSERTI128rr VR256:$src1, VR128:$src2,
8194 (INSERT_get_vinsert128_imm VR256:$ins))>;
8195 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
8197 (VINSERTI128rr VR256:$src1, VR128:$src2,
8198 (INSERT_get_vinsert128_imm VR256:$ins))>;
8199 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
8201 (VINSERTI128rr VR256:$src1, VR128:$src2,
8202 (INSERT_get_vinsert128_imm VR256:$ins))>;
8204 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
8206 (VINSERTI128rm VR256:$src1, addr:$src2,
8207 (INSERT_get_vinsert128_imm VR256:$ins))>;
8208 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
8209 (bc_v4i32 (memopv2i64 addr:$src2)),
8211 (VINSERTI128rm VR256:$src1, addr:$src2,
8212 (INSERT_get_vinsert128_imm VR256:$ins))>;
8213 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
8214 (bc_v16i8 (memopv2i64 addr:$src2)),
8216 (VINSERTI128rm VR256:$src1, addr:$src2,
8217 (INSERT_get_vinsert128_imm VR256:$ins))>;
8218 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
8219 (bc_v8i16 (memopv2i64 addr:$src2)),
8221 (VINSERTI128rm VR256:$src1, addr:$src2,
8222 (INSERT_get_vinsert128_imm VR256:$ins))>;
8225 //===----------------------------------------------------------------------===//
8226 // VEXTRACTI128 - Extract packed integer values
8228 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
8229 (ins VR256:$src1, i8imm:$src2),
8230 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8232 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
8234 let neverHasSideEffects = 1, mayStore = 1 in
8235 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
8236 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
8237 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8240 let Predicates = [HasAVX2] in {
8241 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8242 (v2i64 (VEXTRACTI128rr
8243 (v4i64 VR256:$src1),
8244 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8245 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8246 (v4i32 (VEXTRACTI128rr
8247 (v8i32 VR256:$src1),
8248 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8249 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8250 (v8i16 (VEXTRACTI128rr
8251 (v16i16 VR256:$src1),
8252 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8253 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8254 (v16i8 (VEXTRACTI128rr
8255 (v32i8 VR256:$src1),
8256 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8258 def : Pat<(alignedstore (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
8259 (iPTR imm))), addr:$dst),
8260 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8261 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8262 def : Pat<(alignedstore (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
8263 (iPTR imm))), addr:$dst),
8264 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8265 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8266 def : Pat<(alignedstore (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
8267 (iPTR imm))), addr:$dst),
8268 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8269 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8270 def : Pat<(alignedstore (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
8271 (iPTR imm))), addr:$dst),
8272 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8273 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8276 //===----------------------------------------------------------------------===//
8277 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
8279 multiclass avx2_pmovmask<string OpcodeStr,
8280 Intrinsic IntLd128, Intrinsic IntLd256,
8281 Intrinsic IntSt128, Intrinsic IntSt256> {
8282 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
8283 (ins VR128:$src1, i128mem:$src2),
8284 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8285 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
8286 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
8287 (ins VR256:$src1, i256mem:$src2),
8288 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8289 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
8291 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8292 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8293 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8294 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8295 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8296 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8297 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8298 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8301 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8302 int_x86_avx2_maskload_d,
8303 int_x86_avx2_maskload_d_256,
8304 int_x86_avx2_maskstore_d,
8305 int_x86_avx2_maskstore_d_256>;
8306 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8307 int_x86_avx2_maskload_q,
8308 int_x86_avx2_maskload_q_256,
8309 int_x86_avx2_maskstore_q,
8310 int_x86_avx2_maskstore_q_256>, VEX_W;
8313 //===----------------------------------------------------------------------===//
8314 // Variable Bit Shifts
8316 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8317 ValueType vt128, ValueType vt256> {
8318 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8319 (ins VR128:$src1, VR128:$src2),
8320 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8322 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8324 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8325 (ins VR128:$src1, i128mem:$src2),
8326 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8328 (vt128 (OpNode VR128:$src1,
8329 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8331 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8332 (ins VR256:$src1, VR256:$src2),
8333 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8335 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8337 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8338 (ins VR256:$src1, i256mem:$src2),
8339 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8341 (vt256 (OpNode VR256:$src1,
8342 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8346 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8347 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8348 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8349 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8350 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8352 //===----------------------------------------------------------------------===//
8353 // VGATHER - GATHER Operations
8354 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8355 X86MemOperand memop128, X86MemOperand memop256> {
8356 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8357 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8358 !strconcat(OpcodeStr,
8359 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8361 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8362 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8363 !strconcat(OpcodeStr,
8364 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8365 []>, VEX_4VOp3, VEX_L;
8368 let mayLoad = 1, Constraints
8369 = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
8371 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8372 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8373 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8374 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8375 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8376 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8377 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8378 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;